An intelligent extraction method for complex types of particle-filled porous media

Abstract

The present invention discloses a method for intelligently extracting complex types of particle-filled porous media, comprising the following steps: first, establishing a data set and training a UNeXt model; second, using the UNeXt model to segment porous media images; third, extracting behavioral complexity and porosity; and fourth, extracting original complexity. The present invention trains the UNeXt model by establishing a data set to achieve the purpose of segmenting porous media images, and then extracts the behavioral complexity, original complexity, and porosity of the porous media through different complexity extraction codes. The data can be used to more accurately describe and analyze the pore structure of the porous media, providing a new, efficient and accurate method for the study of porous media.

Description

Technical Field

[0001] The present invention relates to the technical field of porous media performance research, and in particular to a complex type intelligent extraction method for particle-filled porous media. Background Art

[0002] Porous media play an extremely important role in industrial production and natural resource development. The pore structure inside porous media has a crucial influence on the transfer, adsorption and reaction between fluids and solids. Therefore, the study and analysis of the fractal characteristics of porous media can distinguish the adsorption, transport and reaction performance of porous media.

[0003] At present, the research on the pore structure of porous media is only at the stage of fractal theory and network model. The pore structure of porous media is simulated by network model, and the characteristics and properties of the pore structure are described in combination with fractal theory to facilitate the analysis and application of the performance of porous media. However, due to the extremely complex pore structure of porous media, traditional methods that only use structural simulation methods cannot accurately simulate the complete pore structure, resulting in certain deviations in the judgment of porous media performance. At the same time, the method of combining fractal theory with network model cannot accurately derive the complexity of porous media, which affects the research on porous media. Therefore, the present invention proposes a complex type intelligent extraction method for particle-filled porous media to solve the problems existing in the prior art. Summary of the Invention

[0004] In response to the above problems, the purpose of the present invention is to propose an intelligent extraction method for complex types of particle-filled porous media. The intelligent extraction method for complex types of particle-filled porous media achieves the purpose of segmenting porous media images by training the UNeXt model through establishing a data set, and then realizes the extraction of behavioral complexity, original complexity and porosity of porous media through different complexity extraction codes. The data can be used to more accurately describe and analyze the pore structure of porous media, providing a new, efficient and accurate method for the study of porous media.

[0005] To achieve the purpose of the present invention, the present invention is implemented by the following technical solution: a complex type intelligent extraction method of particle-filled porous media, comprising the following steps:

[0006] Step 1: Dataset establishment and model training: Establish a data set and divide it into a training set and a validation set. Use the training set and validation set to train and validate the UNeXt model to obtain the trained UNeXt model.

[0007] Step 2: Porous media image processing: The porous media image is input into the trained UNeXt model for processing, and the UNeXt model outputs a predicted segmentation image;

[0008] Step 3: Behavioral complexity and porosity extraction: The segmented image is input into a Python-based edge extraction code for solid phase feature extraction. The number of all solid phase pixels and the number of closed areas in the image are obtained, and the K-value clustering analysis algorithm is used for intelligent classification. Subsequently, the intelligent extraction algorithm combined with fractal topology theory is used to calculate the scaling gap P, scaling coverage F, and porosity φ;

[0009] Step 4: Extraction of original complexity: Input the segmented image into the Python-based original complexity line extraction code to extract the original complexity, obtain the information of the pore structure, determine the original complexity characteristics, and accurately characterize them.

[0010] A further improvement is that the data set established in step 1 includes the original image and binary labels of the porous medium, wherein Label-me annotation software is used to manually annotate the original image of the porous medium and to perform manual solid phase segmentation annotation on the porous medium labels, and then the annotated porous medium labels are binarized to obtain binary labels.

[0011] A further improvement is that the training set divided in step 1 accounts for 80% of the data set, and the validation set accounts for 20% of the data set. After the division, the porous medium image is subjected to resolution unification processing, and the image resolution is unified into 256*256.

[0012] A further improvement is that the solid phase feature extraction in step three is specifically to input the segmented image into the Python-based edge extraction code, use the Numpy library, convert the three channels to grayscale, set the threshold, convert the image into a matrix of 0 and 1, and use the Find Contours function to extract the edge with the number 1. Finally, use the Contour Area function to obtain the number of pixels of all solid phases in the closed area and the total number of primary, secondary, and tertiary solid phases.

[0013] A further improvement is that in the step three, the intelligent classification specifically first randomly selects K = 3 samples from the data set as cluster centers, then calculates the distance between the samples in all data sets and the cluster centers, and divides each sample into the cluster where the cluster center closest to it is located. For the new cluster, the new cluster center of each cluster is calculated and updated, and the above is repeated until the cluster center does not move, that is, the intelligent classification of all extracted solid phases is completed, and the number of pixels and connected domains of each solid phase is obtained.

[0014] A further improvement is that the scaled clearance P, scaled coverage F and porosity φ in step 3 are calculated by the following formulas:

[0015]

[0016]

[0017]

[0018] Where M1, M2 and M3 represent the number of primary, secondary and tertiary solid phases, respectively; S1, S2 and S3 represent the total number of primary, secondary and tertiary solid phase pixels, respectively; S0 represents the total number of solid phase pixels; and S represents the total number of pixels in the fractal porous media image.

[0019] Further improvements are as follows: Step 4 specifically involves first converting the image information of the porous medium into digital information, and by setting a threshold, screening the edge information of the secondary solid phase and the tertiary solid phase, and then using the Draw Contours function to assign all pixel points in the connected domain of the secondary solid phase and the tertiary solid phase to 0, retaining only the primary solid phase image information and outputting it, extracting the pore structure size, quantity, distribution characteristics and degree of connectivity of the primary solid phase to determine the original complexity characteristics of the porous medium.

[0020] The beneficial effects of the present invention are as follows: the present invention achieves the purpose of segmenting porous media images by establishing a data set to train the UNeXt model, and then realizes the extraction of the behavioral complexity, original complexity and porosity of the porous media through different complexity extraction codes. The data can be used to more accurately describe and analyze the pore structure of the porous media, providing a new, efficient and accurate method for the study of porous media. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Flow chart of the method of the present invention.

[0022] Figure 2 This is a framework diagram of the extraction method of the present invention.

[0023] Figure 3 Schematic diagram of the data set of the present invention.

[0024] Figure 4 Schematic diagram of pore-solid adhesion of particle-filled porous media of the present invention.

[0025] Figure 5 Schematic diagram of the extraction process of the original complexity of the present invention. DETAILED DESCRIPTION

[0026] In order to deepen the understanding of the present invention, the present invention will be further described in detail below with reference to the examples. The examples are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0027] according to Figure 1-Figure 5 As shown, this embodiment provides a complex type intelligent extraction method for particle-filled porous media, comprising the following steps:

[0028] Step 1: Dataset Establishment and Model Training: A dataset was established and divided into a training set and a validation set. The training set accounted for 80% of the dataset and contained 4,305 images of particle-filled porous media. The validation set accounted for 20% of the dataset and contained 1,077 images of particle-filled porous media. After the division, the porous media images were uniformly processed to a resolution of 256*256 to facilitate UNeXt model training. The UNeXt model was then trained and validated using the training and validation sets to obtain the trained UNeXt model.

[0029] The dataset includes the original image and binary labels of porous media, with a total of 5382 images of different scaled gaps P, different scaled coverage F and different porosities. The particle-filled porous medium image is manually annotated using Label-me annotation software, and the porous medium labels are manually segmented and annotated. This can segment all solid phases in the particle-filled porous medium.

[0030] The labeled porous media labels are then binarized to obtain binary labels. Specifically, based on the Python programming language, the Cv2.Split function is used to separate channels and distinguish light and dark colors (light RGB 25525525, dark RGB 5916254); first, the foreground pixel value is obtained (in this case, it is a grayscale image, 0 represents pure black, and 255 represents pure white); then the background pixel value is obtained and a binary image is output. The data set is attached to the manual. Figure 3 shown.

[0031] Step 2: Processing the porous media image: Input the porous media image into the trained UNeXt model to segment the adhered solid phase in the image, obtain independent closed intervals, and output the predicted segmentation image;

[0032] To verify the segmentation effect of the UNeXt model, we first generate particle-filled porous media images based on the four-parameter random growth QSGS algorithm. The pore-solid adhesion is shown in the attached manual. Figure 4 As shown in the figure, the darker colors represent the solid phase and the lighter colors represent the pore phase. This indicates that the solid phase is highly adhered, which is also a result of behavioral complexity. Therefore, according to fractal topology theory, segmenting the adhered solid phase in the pores is a key step in extracting complex types.

[0033] Step 3: Behavioral complexity and porosity extraction: The segmented image is input into the Python-based edge extraction code for solid phase feature extraction, the number of all solid phase pixels and the number of closed areas in the image are obtained, and the K value clustering analysis algorithm is used for intelligent classification. Then, the intelligent extraction algorithm is combined with fractal topology theory to calculate the scaling gap P, scaling coverage F and porosity.

[0034] According to the fractal topology theory, the extraction of the behavioral complexity of fractal porous media requires two important parameters: scaling gap P and scaling coverage F. Scaling gap P is the continuous scaling object G(l i ) and G(l i+1 ) is a characteristic of the scale ratio, and the scale is obtained by extracting the area of ​​each solid phase in the particle-filled porous medium; F is the continuous scaling object G(l i ) and G(l i+1 ), by extracting the number of all connected domains in the particle-filled porous medium, the number of continuously scaled objects in the particle-filled porous medium is obtained;

[0035] Solid phase feature extraction involves inputting the segmented image into a Python-based edge extraction code, using the Numpy library to convert the three channels to grayscale, setting a threshold to convert the image into a matrix of 0s and 1s, and using the Find Contours function to extract edges with a value of 1. Finally, the Contour Area function is used to obtain the number of all solid phase pixels within the enclosed area and the total number of primary, secondary, and tertiary solid phases.

[0036] After obtaining the solid phase information of the particle-filled porous medium, the primary, secondary, and tertiary solid phases are intelligently identified. Since the area and quantity of the primary, secondary, and tertiary solid phases are quite different, the K-means clustering analysis algorithm based on Python code is used for intelligent classification. Specifically, K=3 samples are randomly selected from the data set as cluster centers. Then, the distance between the samples and the cluster centers in all data sets is calculated, and each sample is divided into the cluster with the closest cluster center. For each new cluster, the new cluster center is calculated and updated. The above process is repeated until the cluster center does not move, thus completing the intelligent classification of all extracted solid phases and obtaining the number of pixels and connected domains of each solid phase.

[0037] Among them, the scaled gap P, scaled coverage F and porosity Calculated by the following formulas

[0038]

[0039]

[0040]

[0041] Where M1, M2 and M3 represent the number of primary, secondary and tertiary solid phases, respectively; S1, S2 and S3 represent the total number of primary, secondary and tertiary solid phase pixels, respectively; S0 represents the total number of solid phase pixels; and S represents the total number of pixels in the fractal porous media image.

[0042] Step 4: Extracting the original complexity: Input the segmented image into the Python-based original complexity line extraction code to extract the original complexity, obtain the pore structure information, determine the original complexity characteristics, and accurately characterize them;

[0043] Since the behavioral complexity extraction of particle-filled porous media has been completed in step three, the K-means clustering analysis algorithm has classified the various solid phases in the particle-filled porous media. Therefore, to extract the original complexity of the particle-filled porous media, it is only necessary to extract the first-level solid phase information and features. Specifically, the porous medium image information is first converted into digital information, and the edge information of the secondary and tertiary solid phases is screened by setting a threshold. Then, the DrawContours function is used to assign all pixel points in the connected domain of the secondary and tertiary solid phases to 0, and only the first-level solid phase image information is retained and output. The pore structure size, quantity, distribution characteristics, and degree of connectivity of the first-level solid phase are extracted, as shown in the attached manual. Figure 5 As shown, the original complexity characteristics of porous media can be determined and accurately characterized.

[0044] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A complex type intelligent extraction method for particle-filled porous media, characterized in that: The following steps are involved: Step 1: Dataset establishment and model training: Establish a data set and divide it into a training set and a validation set. Use the training set and validation set to train and validate the UNeXt model to obtain the trained UNeXt model. Step 2: Porous media image processing: The porous media image is input into the trained UNeXt model for processing, and the UNeXt model outputs a predicted segmentation image; Step 3: Behavioral complexity and porosity extraction: The segmented image is input into a Python-based edge extraction code for solid phase feature extraction. The number of all solid phase pixels and the number of closed areas in the image are obtained, and the K-value clustering analysis algorithm is used for intelligent classification. Subsequently, the intelligent extraction algorithm combined with fractal topology theory is used to calculate the scaling gap P, scaling coverage F, and porosity φ; The solid phase feature extraction specifically involves inputting the segmented image into the Python-based edge extraction code, using the Numpy library, converting the three channels to grayscale, and converting the image into a matrix of 0s and 1s by setting a threshold. The Find Contours function is then used to extract the edges with the number 1. Finally, the Contour Area function is used to obtain the number of all solid phase pixels within the closed area and the total number of primary, secondary, and tertiary solid phases. The scaled clearance P, scaled coverage F, and porosity φ are calculated by the following formulas: ； ； ； in 、 and represent the number of primary, secondary, and tertiary solid phases, respectively. 、 and Represents the sum of primary, secondary and tertiary solid-phase pixels respectively, represents the sum of all pixels in the solid phase, Represents the total number of pixels of the fractal porous media image; Step 4: Extracting the original complexity: Input the segmented image into the Python-based original complexity line extraction code to extract the original complexity, obtain the pore structure information, determine the original complexity characteristics, and accurately characterize them; Specifically, the image information of the porous medium is first converted into digital information, and the edge information of the secondary solid phase and the tertiary solid phase is screened by setting a threshold. Then, the Draw Contours function is used to assign all pixel points in the connected domain of the secondary solid phase and the tertiary solid phase to 0, retaining only the primary solid phase image information and outputting it, extracting the pore structure size, quantity, distribution characteristics and degree of connectivity of the primary solid phase to determine the original complexity characteristics of the porous medium.

2. The method for intelligently extracting complex types of particle-filled porous media according to claim 1, characterized in that: The data set established in step 1 includes the original image and binary labels of the porous medium, wherein Label-me annotation software is used to manually annotate the original image of the porous medium and perform manual solid phase segmentation annotation on the porous medium labels, and then the annotated porous medium labels are binarized to obtain binary labels.

3. The method for intelligently extracting complex types of particle-filled porous media according to claim 1, characterized in that: The training set divided in step 1 accounts for 80% of the data set, and the validation set accounts for 20% of the data set. After the division, the porous medium image is subjected to resolution unification processing, and the image resolution is unified into 256*256.

4. The method for intelligently extracting complex types of particle-filled porous media according to claim 1, characterized in that: In the step 3, the intelligent classification specifically first randomly selects K=3 samples from the data set as cluster centers, then calculates the distance between the samples in all data sets and the cluster centers, and divides each sample into the cluster with the cluster center closest to it. For the new cluster, the new cluster center of each cluster is calculated and updated, and the above is repeated until the cluster center does not move, that is, the intelligent classification of all extracted solid phases is completed, and the number of pixels and connected domains of each solid phase is obtained.

Images