Three-dimensional rock image simulation identification method based on adaptive network

Abstract

The invention discloses a three-dimensional rock image simulation identification method based on an adaptive network. According to the method, on the basis of a three-dimensional digital rock microstructure image, a probability density map of voxel information is generated through image feature extraction, and then self-adaptive Voronoi grid division is performed on the three-dimensional image by applying Neper software on the basis of the probability density map; the regularity, roundness, edge fitting rate and texture characterization rate of the grid structure are continuously and iteratively optimized through gravity center iteration and gradient descent, a numerical model capable of reflecting a real rock mesoscopic particle structure is finally obtained, and the mineral proportion, the chimeric relation and the texture distribution of a three-dimensional digital image can be well reproduced. According to the method, the real microscopic particle structure of the digital rock image is considered, a program architecture for transition from the three-dimensional digital image to the three-dimensional numerical model is built, the problem that the existing three-dimensional numerical model is difficult to reflect the rock particle structure is solved to a certain extent, and the method has a good application prospect.

Description

Technical Field

[0001] The invention belongs to the field of three-dimensional modeling image recognition, and in particular relates to a three-dimensional rock image simulation recognition method based on an adaptive network. Background Art

[0002] Rock is a geological material with complex macro- and micro-structures. Due to the complexity of its formation process, rock materials contain different micro-structures such as non-uniform mineral particles and micropores. At present, the use of numerical simulation methods to study the macro-mechanical response of rock materials has been proven to be feasible in technical theory, especially in two-dimensional heterogeneous rocks and three-dimensional homogeneous rock materials. However, if the mechanical effects of three-dimensional real rocks with heterogeneity and anisotropy are simulated and analyzed using the method of studying homogeneous materials and two-dimensional models, the simulation results will be distorted. At the same time, the key to numerical simulation calculations is the division of grids, and the quality of grid division will have a great impact on the accuracy of numerical calculation results. Therefore, in order to truly reflect the mechanical response of rock materials in nature, accurately establish the mapping relationship from three-dimensional digital rock images to three-dimensional grid structures, and divide high-quality computational grids for the identified rock microstructure models, it is an urgent link to be solved.

[0003] At present, the modeling methods for three-dimensional heterogeneous rocks mainly include: (1) randomly generating heterogeneous models containing multiphase minerals through algorithms. (2) mapping the voxel information of digital rocks one by one based on hexahedral grids. Although the above two methods can simulate the mechanical response of rock structures to a certain extent, they also have significant shortcomings: (1) randomly generated multiphase mineral particle rocks are difficult to truly reflect the texture characteristics of rocks and mineral mosaic relationships. (2) regular hexahedral grids are not only extremely computationally expensive, but the simulation results are also difficult to simulate the real rock mechanics mechanism. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention provides a three-dimensional rock image simulation and recognition method based on an adaptive network. Starting from reproducing the real texture structure and mineral mosaic relationship of three-dimensional digital rock, a rock modeling method at a three-dimensional microscopic heterogeneous scale is developed, and an adaptive three-dimensional grid division system based on rock image information is established to provide a reliable numerical model for the three-dimensional rock microscopic model.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a three-dimensional rock image simulation recognition method based on an adaptive network, comprising the following steps:

[0006] S1, performing three-dimensional superposition processing on the two-dimensional digital rock section image to obtain a three-dimensional digital rock grayscale image;

[0007] S2, extracting features from the three-dimensional digital rock grayscale image, obtaining seed points generated based on the probability density map, and then performing preliminary division of the three-dimensional Voronoi polyhedron grid;

[0008] S3, according to the geometric information of the three-dimensional Voronoi polyhedron mesh, the seed point of each polyhedron is shifted at the centroid; the centroids of all polyhedrons are used as seed points, and the Voronoi mesh subdivision is continuously iterated and optimized to obtain the Voronoi polyhedron after centroid iteration;

[0009] S4, establishing an objective function for the edge fit degree of the Voronoi mesh structure and voxel feature information for the three-dimensional Voronoi polyhedron mesh after the centroid iteration;

[0010] S5. Use the gradient descent method to find the size and direction of the objective function approaching the minimum point, and establish a theoretical formula for the size and direction of the gradient descent; according to the theoretical formula, offset each seed point of the Voronoi grid structure in the direction of the gradient descent, and the offset is the corresponding gradient descent value, and continuously iterate and optimize until the offset of the seed point reaches a stable value; perform phase separation on the three-dimensional Voronoi polyhedron grid structure according to the voxel value corresponding to each seed point, and assign colors to different polyhedrons;

[0011] S6. Return to step S3 and execute until a preset number of iterations is reached to obtain a three-dimensional Voronoi polyhedron mesh structure after iterative optimization.

[0012] Furthermore, the aforementioned step 1 includes the following sub-steps:

[0013] S1.1. Obtain a cubic rock sample and scan it to obtain a two-dimensional digital rock section image;

[0014] S1.2. Based on MATLAB image processing technology, firstly, the pixel information and RGB color distribution histogram of each image are obtained through imread and imhist functions respectively; then, combined with the color of the mineral components in the image, the interval and segmentation threshold of the R, G, and B values ​​of each mineral component are determined, the mineral components are accurately segmented, and the RGB image is converted into a grayscale image.

[0015] S1.3. Use MATLAB to upgrade the two-dimensional pixel information into three-dimensional voxel information and synthesize it into a three-dimensional digital rock grayscale image.

[0016] Furthermore, the aforementioned step S2 includes the following sub-steps:

[0017] S2.1. Performing edge detection of the mineral components of the three-dimensional digital rock grayscale image using the Canny operator to obtain the boundaries of the particles in each mineral component and obtain edge skeleton features of the three-dimensional digital rock;

[0018] S2.2, calculating the minimum distance from the voxel point in the image to the feature boundary, establishing the probability density function of the voxel point, and generating the probability density map of the three-dimensional digital rock image;

[0019] S2.3, scattering the initial Voronoi grid seed points, and screening and eliminating the number of seed points according to the probability density map;

[0020] S2.4. Based on the Linux system Neper crystal generation program, the three-dimensional image is initially divided into Voronoi grids according to the seed point coordinates.

[0021] Furthermore, in the aforementioned step S2.2, the probability density function of the voxel point is as follows:

[0022]

[0023] Among them, dist(i,j,k) is the set of minimum distances from all voxel points to the feature boundary.

[0024] Furthermore, in the aforementioned step S3, the centroid shift is specifically as follows: first, all faces of each Voronoi polyhedron are divided into triangular faces using the convhull function; secondly, each triangular face of each polyhedron and its corresponding seed point form a tetrahedron, the volume and centroid of the tetrahedron are calculated, and the centroid coordinates of the polyhedron are calculated using the centroid formula of the combination, and the above steps are executed repeatedly to obtain the centroid coordinates of all polyhedrons.

[0025] Furthermore, the aforementioned barycenter coordinate calculation formula is as follows:

[0026]

[0027] in, is the barycentric coordinates of a polyhedron, V(x i ) is the volume of a tetrahedron in the polyhedron, x i are the centroid coordinates of the tetrahedron.

[0028] Furthermore, in the aforementioned step S4, the objective function is established as follows: first, the discreteness of the voxel points in each polyhedron is obtained, and the calculation formula is the square of the difference between each voxel point in a polyhedron and the average value of the voxels in the polyhedron, which is defined as the local error of the Voronoi grid structure; secondly, the local errors in all polyhedrons are added to obtain the global error of the entire Voronoi grid structure, which is the objective function of gradient descent, as shown in the following formula:

[0029]

[0030] Among them, P(x) represents the voxel value of the image, Represents the three-dimensional Voronoi polyhedron V i The average value of the voxels contained in N i Represents a polyhedron V i The number of voxels contained in Represents a polyhedron V i The sum of the voxels contained in .

[0031] Furthermore, in the aforementioned step S5, the gradient descent method is used to find the magnitude and direction of the gradient descent of the objective function, and a theoretical formula for the magnitude and direction of the gradient descent is established. The gradient descent method is as follows:

[0032]

[0033] Among them, this formula indicates that the i+1th point is determined by the i-th point and gradient before it, and η represents the magnitude of the gradient at the i-th point. Indicates the gradient direction at the i-th point. The minus sign indicates the opposite direction of the gradient and is used for gradient descent to solve the minimum value.

[0034] Furthermore, in the aforementioned step S5, a theoretical formula for the magnitude and direction of the gradient descent is established, and the specific steps include S5.1 to S5.3:

[0035] S5.1. First, the partial derivative of the objective function that represents the degree of fit of each polygon edge is obtained as follows:

[0036]

[0037] Where l represents the seed point i and its adjacent seed point J i The collection of V l S5.2, based on the interrelated nature of Voronoi polyhedra, consider the envelope seed points around each polyhedron, and use the Leibniz formula as follows to expand the formula in step S4.1:

[0038]

[0039] Among them, D t is a smooth two-dimensional interval defined at time t, g(x,t) is defined on D t A function on For region D t The boundary of The velocity vector is expressed as dx / dt, n is the normal vector of the point on the boundary, pointing outward, and ds is the unit arc length on the boundary;

[0040] S5.3. Use the properties of the perpendicular bisectors of the Voronoi polyhedron to eliminate the constant term, and finally obtain the gradient size and direction of the edge fit as follows:

[0041]

[0042] Among them, σ i represents the standard deviation of the voxels within the i-th Voronoi polyhedron, σ j represents the standard deviation of the voxels in the adjacent polyhedron, J i Represents the seed point x i The number of adjacent seed points, k represents the number of iterations, v i represents the volume of the i-th Voronoi polyhedron.

[0043] Furthermore, in the aforementioned step S3, iterative optimization of Voronoi mesh division is performed, and all information of the Voronoi mesh is output as tess and obj files.

[0044] Compared with the prior art, the beneficial technical effects of the present invention using the above technical solution are as follows:

[0045] 1. The present invention provides a method for adaptively performing a rock image in a three-dimensional space.

[0046] 2. The grid structure modeling method reproduces the texture characteristics and mineral mosaic relationship of three-dimensional rocks well, and can better reflect the spatial distribution characteristics of rock particle materials than the traditional two-dimensional heterogeneous rock modeling method.

[0047] 3. The present invention reconstructs the real mesostructure of rock based on three-dimensional rock images, which can better reflect the mesostructure of natural rocks than the mesostructure model generated by traditional random functions, and provides reliable model data for the numerical analysis of relevant rock materials.

[0048] 4. The present invention uses a three-dimensional Voronoi grid structure instead of a hexahedral grid structure, which reduces the computational cost of establishing a hexahedral grid by one-to-one correspondence of pixels or voxel points in the traditional method, and overcomes the distortion of the hexahedral grid in simulating the mechanical behavior of rock materials; the three-dimensional Voronoi grid has the properties of convex polygons, empty circles, minimum distance, etc. in terms of geometric characteristics. The regular polyhedron is closer to the microscopic particle structure of the rock, and the corresponding public face group can better reflect the texture surface of the three-dimensional rock.

[0049] 5. The present invention iteratively optimizes the Voronoi grid structure based on the center of gravity shift and gradient descent method, clarifies the center of gravity shift principle based on three-dimensional Voronoi polygons, derives the gradient descent theoretical formula based on voxel point information, and better reproduces the three-dimensional rock image in terms of mineral proportion, mosaic relationship and texture distribution.

[0050] 6. The present invention is applicable to a variety of rock materials, as well as concrete, porous structures and other materials, and has a wide range of application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a flow chart of the method of the present invention;

[0052] Figure 2 It is a grayscale image of the two-dimensional cross-section image of the cubic basalt rock sample in Example 1 of the present invention after threshold segmentation. There are a total of 175 images, 3 of which are presented here. The physical size of the image is 12 cm×12 cm, and the digital image size is 175 pixels×175 pixels;

[0053] Figure 3 It is a three-dimensional digital rock grayscale image obtained by superimposing and merging two-dimensional grayscale images. The size of the digital image is 175 voxels × 175 voxels × 175 voxels;

[0054] Figure 4 It is a schematic diagram of using the Canny operator to extract the edge skeleton features of a three-dimensional grayscale image.

[0055] Figure 5 It is a distribution map of seed points in spatial coordinates scattered according to probability density information.

[0056] Figure 6 It is a three-dimensional Voronoi grid model diagram for initial grid generation.

[0057] Figure 7 This is the Voronoi grid model diagram after the center of gravity shift iteration, where (a), (b), and (c) are the models after the 1st, 5th, and 10th iterations, respectively.

[0058] Figure 8This is the Voronoi grid model diagram after gradient descent iteration, where (a), (b), and (c) are the models after the 1st, 3rd, and 5th iterations respectively.

[0059] Fig. 9 The three-dimensional Voronoi grid model diagram of the present invention in different mineral phases, wherein (a) is the overall three-dimensional Voronoi grid model, (b) is biotite, (c) is quartz, and (d) is feldspar.

[0060] Fig.10 The voxel model and the three-dimensional Voronoi grid model diagram of the three-dimensional rock image finally presented by the present invention. (a) is the RGB color voxel file presented in the Figure window in MATLAB; (b) is the grayscale voxel file after threshold segmentation; (c) is the stl file presented in the multi-physics field coupling software Comsol. DETAILED DESCRIPTION

[0061] In order to better understand the technical content of the present invention, specific embodiments are given and described as follows in conjunction with the accompanying drawings.

[0062] Various aspects of the invention are described herein with reference to the accompanying drawings, in which many illustrative embodiments are shown. The embodiments of the invention are not limited to those described in the accompanying drawings. It should be understood that the invention is implemented by any of the various concepts and embodiments described above, as well as the concepts and embodiments described in detail below, because the concepts and embodiments disclosed in the invention are not limited to any implementation. In addition, some aspects disclosed in the invention may be used alone or in any appropriate combination with other aspects disclosed in the invention.

[0063] refer to Figure 1 The present invention provides a three-dimensional rock image simulation recognition method based on an adaptive network, comprising the following steps:

[0064] S1, performing three-dimensional superposition processing on the two-dimensional digital rock section image to obtain a three-dimensional digital rock grayscale image;

[0065] S2, extracting features from the three-dimensional digital rock grayscale image, obtaining seed points generated based on the probability density map, and then performing preliminary division of the three-dimensional Voronoi polyhedron grid;

[0066] S3, according to the geometric information of the three-dimensional Voronoi polyhedron mesh, the seed point of each polyhedron is shifted at the centroid; the centroids of all polyhedrons are used as seed points, and the Voronoi mesh subdivision is continuously iterated and optimized to obtain the Voronoi polyhedron after centroid iteration;

[0067] S4, establishing an objective function for the edge fit degree of the Voronoi mesh structure and voxel feature information for the three-dimensional Voronoi polyhedron mesh after the centroid iteration;

[0068] S5. Use the gradient descent method to find the size and direction of the objective function approaching the minimum point, and establish a theoretical formula for the size and direction of the gradient descent; according to the theoretical formula, offset each seed point of the Voronoi grid structure in the direction of the gradient descent, and the offset is the corresponding gradient descent value, and continuously iterate and optimize until the offset of the seed point reaches a stable value; perform phase separation on the three-dimensional Voronoi polyhedron grid structure according to the voxel value corresponding to each seed point, and assign colors to different polyhedrons;

[0069] S6. Return to step S3 and execute until a preset number of iterations is reached to obtain a three-dimensional Voronoi polyhedron mesh structure after iterative optimization.

[0070] As a preferred embodiment of the three-dimensional rock image simulation recognition method based on an adaptive network of the present invention, step S1 includes the following sub-steps:

[0071] S1.1. Obtain a cubic rock sample and scan it to obtain a two-dimensional digital rock section image. Specifically: select a cubic granite sample with a rock size of 12cm×12cm×12cm, and use a scanner to scan a series of two-dimensional digital rock slice images with a slice image size of 12cm×12cm. Use MATLAB's image processing technology to perform RGB three-channel multi-threshold segmentation on the surface texture image of each cut surface of the rock sample, and use different grayscale values ​​to represent different mineral components, including black-biotite, gray-quartz, white-feldspar, and some of the two-dimensional slice images after threshold segmentation are as follows Figure 2 shown.

[0072] S1.2. Based on MATLAB image processing technology, firstly, the pixel information and RGB color distribution histogram of each image are obtained through imread and imhist functions respectively; then, combined with the color of the mineral components in the image, the interval and segmentation threshold of the R, G, and B values ​​of each mineral component are determined, the mineral components are accurately segmented, and the RGB image is converted into a grayscale image.

[0073] S1.3, using MATLAB, the two-dimensional pixel information is upgraded to three-dimensional voxel information and synthesized into a three-dimensional digital rock grayscale image, such as Figure 3 shown.

[0074] As a preferred embodiment of the three-dimensional rock image simulation recognition method based on an adaptive network of the present invention, step S2 includes the following sub-steps:

[0075] S2.1. Perform edge detection of the mineral components of the three-dimensional digital rock grayscale image using the Canny operator to obtain the boundaries of the particles in each mineral component and obtain the edge skeleton features of the three-dimensional digital rock. The skeleton features are formed by the accumulation of red voxel points. Figure 4 shown.

[0076] S2.2. Calculate the minimum distance from the voxel point in the image to the feature boundary, establish the probability density function of the voxel point, and generate the probability density map of the three-dimensional digital rock image; the probability density function of the voxel point is as follows:

[0077]

[0078] Among them, dist(i,j,k) is the set of minimum distances from all voxel points to the feature boundary.

[0079] S2.3, scatter the initial Voronoi grid seed points, the initial number of seed points is 30,000, and screen and eliminate the number of seed points according to the probability density map. The steps of screening and eliminating seed points are to compare the acceptance probability of each seed point with a random number less than 1 to determine whether to accept it, so as to retain more seed points on the feature boundary and eliminate some seed points inside the same particles. The number of seed points after elimination is 21,001, and the distribution of seed points is as follows: Figure 5 shown.

[0080] S2.4, based on the Neper crystal generation program of the Linux system, the three-dimensional image is initially divided into Voronoi grids according to the seed point coordinates; and all the information of the Voronoi grid is output as tess, obj, and stl files. The command flow of grid division (Formula 2) and image drawing (Formula 3) is as follows. The result after initial grid division is as follows Figure 6 shown.

[0081] neper-Tn 21001-domain"cube(175,175,175)"-id 1-format stl,obj,tess-morphooptiini

[0082] "coo:file(coo_30000_21001),weight:0"-morpho voronoi-o test

[0083] Among them, -T is the subcommand that calls the mosaic module, -n is the number of Voronoi grids, -domain is the grid division area, -id is the file number, -format is the output format of the grid file, -morphooptini is the Voronoi morphology and spatial arrangement, -morpho is the mosaic algorithm, and -o is the file naming format.

[0084] neper-V test.tess-datacellcol id-print test_imag

[0085] Among them, -V is the subcommand that calls the visualization module, -datacellcol is the color of the polyhedron, and -print is the naming format of the output image.

[0086] As a preferred method of three-dimensional rock image simulation recognition based on an adaptive network of the present invention, step S3 is specifically: based on MATLAB, all geometric information of the Voronoi polyhedron grid is read, the seed point of each polyhedron is subjected to a gravity center offset, each triangular face of each polyhedron and its corresponding seed point form a tetrahedron, the volume and gravity center of the tetrahedron are calculated, the gravity center coordinates of the polyhedron are calculated using the gravity center formula of the combination, and the above steps are repeated to obtain the gravity center coordinates of all polyhedrons; the gravity center offset is based on the following formula:

[0087]

[0088] in, is the barycentric coordinates of a polyhedron, V(x i ) is the volume of a tetrahedron in the polyhedron, x i are the centroid coordinates of the tetrahedron.

[0089] The centroids of all polyhedrons are used as seed points, and the Voronoi mesh generation is continuously iterated and optimized until the morphological structure of the polyhedron mesh hardly changes. For this embodiment, the number of iterations is 10, and the Voronoi mesh structures after the 1st, 5th, and 10th iterations are as follows: Figure 7 As shown in (a), (b) and (c);

[0090] As a preferred method of three-dimensional rock image simulation and recognition based on an adaptive network of the present invention, in step S4, the specific steps of establishing the objective function are as follows: first, the discrete degree of the voxel points in each polyhedron is obtained, and the calculation formula is the square of the difference between each voxel point in a polyhedron and the average value of the voxels in the polyhedron, which is defined as the local error of the Voronoi grid structure; secondly, the local errors in all polyhedrons are added to obtain the global error of the entire Voronoi grid structure, which is the objective function of gradient descent, as shown in the following formula:

[0091]

[0092] Among them, P(x) represents the voxel value of the image, Represents the three-dimensional Voronoi polyhedron V i The average value of the voxels contained in N i Represents a polyhedron V i The number of voxels contained in Represents a polyhedron V i The sum of the voxels contained in .

[0093] As a preferred method of three-dimensional rock image simulation and recognition based on an adaptive network of the present invention, in step S4, the gradient descent theory formula is derived as follows: first, the partial derivative of the objective function characterizing the edge fit of each polygon is obtained, and secondly, in view of the interrelated nature of the Voronoi polyhedrons, the envelope seed points around each polyhedron are considered, and the Leibniz formula is used to expand the above formula, and then the properties of the perpendicular bisectors of the Voronoi polyhedrons are used to eliminate the constant term, and finally the gradient size and direction of the edge fit are obtained.

[0094] The gradient descent method is as follows:

[0095]

[0096] Among them, this formula indicates that the i+1th point is determined by the i-th point and gradient before it, and η represents the magnitude of the gradient at the i-th point. Represents the gradient direction at the i-th point. The minus sign indicates the opposite direction of the gradient and is used for gradient descent to solve the minimum value.

[0097] As a preferred embodiment of the three-dimensional rock image simulation recognition method based on an adaptive network of the present invention, step S4 specifically comprises the following steps:

[0098] S4.1. First, the partial derivative of the objective function that represents the degree of fit of each polygon edge is obtained as follows:

[0099]

[0100] Where l represents the seed point i and its adjacent seed point J i The collection of V l S4.2, based on the interrelated nature of Voronoi polyhedra, the envelope seed points around each polyhedron are considered, and the Leibniz formula is used as follows to expand the formula in step S4.1:

[0101]

[0102] Among them, D t is a smooth two-dimensional interval defined at time t, g(x,t) is defined on D t A function on For region D t The boundary of The velocity vector is expressed as dx / dt, n is the normal vector of the point on the boundary, pointing outward, and ds is the unit arc length on the boundary;

[0103] S4.3. Use the properties of the perpendicular bisectors of the Voronoi polyhedron to eliminate the constant term, and finally obtain the gradient size and direction of the edge fit as follows:

[0104]

[0105] Among them, σ i represents the standard deviation of the voxels within the i-th Voronoi polyhedron, σ j represents the standard deviation of the voxels in the adjacent polyhedron, J i Represents the seed point x i The number of adjacent seed points, k represents the number of iterations, v i represents the volume of the i-th Voronoi polyhedron.

[0106] As a preferred method of three-dimensional rock image simulation recognition based on an adaptive network in the present invention, each seed point of the Voronoi grid structure is offset in the gradient descent direction, and the offset is the corresponding gradient descent value. The optimization is iterated continuously until the offset of the seed point reaches a stable value. For this case, the number of iterations of gradient descent is 5, and the Voronoi grid structures after the 1st, 3rd, and 5th iterations are respectively as follows: Figure 8 As shown in (a), (b) and (c);

[0107] The three-dimensional Voronoi grid structure is phase-processed according to the voxel value corresponding to each seed point, and different polyhedrons are colored. The schematic diagram of different mineral phases after coloring is shown as follows Fig. 9 As shown, Fig. 9 (a) shows the overall three-dimensional Voronoi grid model, (b) is biotite, (c) is quartz, and (d) is feldspar;

[0108] Return to step S3 until the preset number of iterations is reached to output the three-dimensional Voronoi grid structure after iterative optimization, and output the three-dimensional Voronoi grid structure of each phase respectively. The output format is tess, obj, and stl files, providing a three-dimensional digital rock adaptive grid model for numerical simulation, such as Fig.10(a) is the presentation result of the RGB voxel file in MATLAB, (b) is the grayscale voxel file after threshold segmentation, and (c) is the stl file presented in the multi-physics field coupling software Comsol.

[0109] In the traditional three-dimensional digital rock modeling method, the randomly generated grid model cannot present the texture characteristics and mineral mosaic relationship of the rock structure, and the hexahedral grid has problems such as high computational cost and distorted simulation results. The method of the present invention better reproduces the mineral proportion, mosaic relationship and texture structure in the rock image, and the three-dimensional adaptive Voronoi grid structure has the potential and value of high-quality simulation of rock mechanical behavior.

[0110] Although the present invention has been described above with preferred embodiments, it is not intended to limit the present invention. A person skilled in the art of the present invention may make various modifications and improvements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the definition of the claims.

Claims

1. A three-dimensional rock image simulation recognition method based on an adaptive network, characterized in that: The steps include: S1, performing three-dimensional superposition processing on the two-dimensional digital rock section image to obtain a three-dimensional digital rock grayscale image; S2, extracting features from the three-dimensional digital rock grayscale image, obtaining seed points generated based on the probability density map, and then performing preliminary division of the three-dimensional Voronoi polyhedron grid; S3, according to the geometric information of the three-dimensional Voronoi polyhedron mesh, the seed point of each polyhedron is shifted at the centroid; the centroids of all polyhedrons are used as seed points, and the Voronoi mesh subdivision is continuously iterated and optimized to obtain the Voronoi polyhedron after centroid iteration; S4, establishing an objective function for the edge fit degree of the Voronoi mesh structure and voxel feature information for the three-dimensional Voronoi polyhedron mesh after the centroid iteration; S5. Use the gradient descent method to find the size and direction of the objective function approaching the minimum point, and establish a theoretical formula for the size and direction of the gradient descent; According to the theoretical formula, each seed point of the Voronoi grid structure is offset in the gradient descent direction. The offset is the corresponding gradient descent value. The optimization is iterated continuously until the offset of the seed point reaches a stable value. According to the voxel value corresponding to each seed point, the three-dimensional Voronoi polyhedron grid structure is phase-processed to assign colors to different polyhedrons. S6. Return to step S3 and execute until a preset number of iterations is reached to obtain a three-dimensional Voronoi polyhedron mesh structure after iterative optimization.

2. The three-dimensional rock image simulation recognition method based on an adaptive network according to claim 1 is characterized in that: Step 1 includes the following sub-steps: S1.

1. Obtain a cubic rock sample and scan it to obtain a two-dimensional digital rock section image; S1.

2. Based on MATLAB image processing technology, firstly, the pixel information and RGB color distribution histogram of each image are obtained through imread and imhist functions respectively; then, combined with the color of the mineral components in the image, the interval and segmentation threshold of the R, G, and B values ​​of each mineral component are determined, the mineral components are accurately segmented, and the RGB image is converted into a grayscale image. S1.

3. Use MATLAB to upgrade the two-dimensional pixel information into three-dimensional voxel information and synthesize it into a three-dimensional digital rock grayscale image.

3. The method for three-dimensional rock image simulation recognition based on an adaptive network according to claim 1, characterized in that: Step S2 includes the following sub-steps: S2.

1. Performing edge detection of the mineral components of the three-dimensional digital rock grayscale image using the Canny operator to obtain the boundaries of the particles in each mineral component and obtain edge skeleton features of the three-dimensional digital rock; S2.2, calculating the minimum distance from the voxel point in the image to the feature boundary, establishing the probability density function of the voxel point, and generating the probability density map of the three-dimensional digital rock image; S2.3, scattering the initial Voronoi grid seed points, and screening and eliminating the number of seed points according to the probability density map; S2.

4. Based on the Linux system Neper crystal generation program, the three-dimensional image is initially divided into Voronoi grids according to the seed point coordinates.

4. The method for three-dimensional rock image simulation recognition based on an adaptive network according to claim 3 is characterized in that: In step S2.2, the probability density function of the voxel point is as follows: Among them, dist(i,j,k) is the set of minimum distances from all voxel points to the feature boundary.

5. The method for three-dimensional rock image simulation recognition based on an adaptive network according to claim 1, characterized in that: In step S3, the gravity center offset is specifically as follows: first, all faces of each Voronoi polyhedron are divided into triangular faces using the convhull function; second, each triangular face of each polyhedron and its corresponding seed point form a tetrahedron, the volume and gravity center of the tetrahedron are calculated, and the gravity center coordinates of the polyhedron are calculated using the gravity center formula of the combination; the above steps are repeated to obtain the gravity center coordinates of all polyhedrons.

6. The method for three-dimensional rock image simulation recognition based on an adaptive network according to claim 4, characterized in that: The calculation formula of the center of gravity coordinates is as follows: in, is the barycentric coordinates of a polyhedron, V(x i ) is the volume of a tetrahedron in the polyhedron, x i are the coordinates of the center of gravity of the tetrahedron.

7. The method for three-dimensional rock image simulation and recognition based on an adaptive network according to claim 1, characterized in that: In step S4, the objective function is established as follows: first, the degree of discreteness of the voxel points in each polyhedron is obtained, and the calculation formula is the square of the difference between each voxel point in a polyhedron and the average value of the voxels in the polyhedron, which is defined as the local error of the Voronoi grid structure; secondly, the local errors in all polyhedrons are added to obtain the global error of the entire Voronoi grid structure, which is the objective function of gradient descent, as shown in the following formula: Where P(x) represents the voxel value of the image, Represents the three-dimensional Voronoi polyhedron V i The average value of the voxels contained in N i Represents a polyhedron V i The number of voxels contained in Represents a polyhedron V i The sum of the voxels contained in .

8. The method for three-dimensional rock image simulation and recognition based on an adaptive network according to claim 1, characterized in that: In step S5, the gradient descent method is used to find the magnitude and direction of the gradient descent of the objective function, and a theoretical formula for the magnitude and direction of the gradient descent is established. The gradient descent method is as follows: Among them, this formula indicates that the i+1th point is determined by the i-th point and gradient before it, and η represents the magnitude of the gradient at the i-th point. Represents the gradient direction at the i-th point, and the minus sign indicates the opposite direction of the gradient, which is used for gradient descent to solve the minimum value.

9. The method for three-dimensional rock image simulation recognition based on an adaptive network according to claim 8, characterized in that: In step S5, a theoretical formula for the magnitude and direction of the gradient descent is established, and the specific steps include S5.1 to S5.3: S5.

1. First, the partial derivative of the objective function that represents the degree of fit of each polygon edge is obtained as follows: Where l represents the seed point i and its adjacent seed point J i The collection of V l The Voronoi polyhedron representing the seed point set; S5.

2. Based on the interrelated nature of Voronoi polyhedra, the envelope seed points around each polyhedron are considered, and the Leibniz formula is used as follows to expand the formula in step S4.1: Among them, D t is a smooth two-dimensional interval defined at time t, g(x,t) is defined on D t A function on For region D t The boundary of The velocity vector is expressed as dx / dt, n is the normal vector of the point on the boundary, pointing outward, and ds is the unit arc length on the boundary; S5.

3. Use the properties of the perpendicular bisectors of the Voronoi polyhedron to eliminate the constant term, and finally obtain the gradient size and direction of the edge fit as follows: Among them, σ i represents the standard deviation of the voxels within the i-th Voronoi polyhedron, σ j represents the standard deviation of the voxels in the adjacent polyhedron, J i Represents the seed point x i The number of adjacent seed points, k represents the number of iterations, v i represents the volume of the i-th Voronoi polyhedron.

10. The method for three-dimensional rock image simulation recognition based on an adaptive network according to claim 1, characterized in that: In step S3, iterative optimization of Voronoi mesh division is performed, and all information of the Voronoi mesh is output as tess and obj files.

Images