Method, system and terminal for generating microstructure image of cement structure based on multi-point statistical modifier semantics and storage medium

Abstract

The application relates to the technical field of image generation, and discloses a cement structure microscopic image generation method and system based on multi-point statistical modifier semantics, a terminal and a storage medium.The method comprises the following steps: acquiring a historical image training set of a cement matrix, carrying out pretreatment to obtain a binary image training set, and performing multi-point statistical training on a target generative adversarial neural network according to the binary image training set to obtain a structure microscopic image generation model; acquiring target image data of a target cement matrix, inputting the target image data into the structure microscopic image generation model, and outputting a cement structure microscopic image of the target cement matrix.The application realizes automatic labeling of the latent space semantics of the generative adversarial neural network by scanning the image through a multi-point statistical algorithm to extract information, improves the controllability of the generated cement microscopic structure image, and fuses the extracted information into a noise vector, so that the efficiency and accuracy of the generated cement microscopic structure image are improved.

Description

Technical Field

[0001] This invention relates to the field of image generation technology, and in particular to a method, system, terminal, and computer-readable storage medium for generating microscopic images of cement structures based on multi-point statistical modification semantics. Background Technology

[0002] Studying the microstructure of cementitious matrices is crucial for accurately predicting the durability and corrosion resistance of cementitious materials. Currently, the most common method for studying the microstructure of cementitious matrices is to directly acquire images of the corresponding microstructure using X-ray computed tomography (XCT). However, this method is costly, with each scan requiring significant expense. With the application of artificial intelligence, trained generative adversarial neural networks (GANs) are being used to generate microscopic images of the structure, replacing the direct scanning method.

[0003] However, due to the lack of data annotation and incomplete experimental data, the reliability and accuracy of the images generated by this method are limited. The main problems with using traditional generative adversarial neural networks to obtain structural micro-images when data annotation is lacking include: (1) the trained model does not intuitively represent the semantics of the latent space, resulting in low controllability of the generated structural micro-images; (2) due to the lack of data annotation, manual labeling is required, resulting in low efficiency in generating structural micro-images; (3) generative adversarial neural networks are more prone to crashing when data is lacking, resulting in low accuracy in generating structural micro-images.

[0004] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0005] The main objective of this invention is to provide a method, system, terminal, and storage medium for generating microscopic images of cement structures based on multi-point statistical modification semantics, aiming to solve the problems of low controllability, efficiency, and accuracy in generating cement microscopic structure images in existing technologies.

[0006] To achieve the above objectives, this invention provides a method for generating microscopic images of cement structures based on multi-point statistical modification semantics. The method includes the following steps: A historical image training set of cement matrix is ​​obtained, and the historical image training set is preprocessed to obtain a binary image training set; Based on the binary image training set, the created target generative adversarial neural network is subjected to multi-point statistical training to obtain a structural microscopic image generation model. Acquire target image data of the target cement matrix, input the target image data into the structural micro-image generation model, and output the cement structure micro-image of the target cement matrix.

[0007] Optionally, the method for generating microscopic images of cement structures based on multi-point statistical modification semantics, wherein obtaining a historical image training set of the cement matrix and preprocessing the historical image training set to obtain a binary image training set specifically includes: A training set of historical images of cement matrix is ​​obtained, wherein the training set of historical images includes multiple sets of historical cement matrix images and corresponding historical cement matrix structure micro images; Pixel values ​​are extracted from all historical cement matrix structure micro images in the historical image training set to obtain image pixel value extraction results, and all image pixel values ​​in the image pixel value extraction results are compared with a preset threshold. If there is a first image pixel value that is greater than or equal to the preset threshold, then the first image pixel value is set to a first value to obtain a first setting result; If there is a second image pixel value that is less than the preset threshold, then the first image pixel value is set to the second value to obtain the second setting result; Multiple binary images are obtained based on the first setting result and the second setting result, and all the binary images are combined with the corresponding historical cement matrix images to obtain a binary image training set.

[0008] Optionally, the method for generating microscopic images of cement structures based on multi-point statistical modification semantics, wherein the step of performing multi-point statistical training on the created target generative adversarial neural network based on the binary image training set to obtain a structural microscopic image generation model specifically includes: Create a target generative adversarial neural network, and input a set of binary image training samples from the binary image training set into the target generative adversarial neural network; Multi-point statistical scanning is performed on the binary image of the training sample to obtain pixel extraction results. Based on the pixel extraction results, binary tree processing is performed to obtain a probability information tree table. The probability information tree table is fused into a vector to obtain a fusion vector. An initial cement structure micro-image is obtained based on the fusion vector. The initial cement structure micro-image and the binary image of the binary image training sample are then subjected to a difference calculation to obtain the difference calculation result. The discriminator loss is calculated based on the difference calculation results to obtain the discriminator loss value. The generator loss is calculated based on the difference calculation results to obtain the generator loss value. The parameters of the target generative adversarial neural network are then corrected based on the discriminator loss and the generator loss. The next set of binary image training samples is input into the target generative adversarial neural network until the training status of the target generative adversarial neural network meets the preset conditions, thus obtaining a trained structural microscopic image generation model.

[0009] Optionally, the method for generating microscopic images of cement structures based on multi-point statistical modification semantics, wherein performing multi-point statistical scanning on the binary images of the binary image training samples to obtain pixel extraction results, and performing binary tree processing based on the pixel extraction results to obtain a probability information tree table, specifically includes: Obtain an image pixel search template, scan the binary image of the binary image training sample pixel by pixel according to the image pixel search template to obtain pixel scanning results, and extract pixels from the pixel scanning results to obtain pixel extraction results; A binary tree is built based on the pixel extraction results. The binary tree is then updated to obtain the target binary tree. The count probability values ​​of all pixels in the target binary tree are calculated to obtain the calculation results. A probability information tree table is then obtained based on the calculation results.

[0010] Optionally, the method for generating microscopic images of cement structures based on multi-point statistical modification semantics, wherein the step of vector fusion of the probability information tree table to obtain a fusion vector, obtaining an initial microscopic image of the cement structure based on the fusion vector, and performing difference calculation on the initial microscopic image of the cement structure and the binary images of the binary image training samples to obtain difference calculation results, specifically includes: Based on the probability information tree table, feature extraction is performed on the historical cement matrix images of the binary image training samples to obtain image feature vectors. Then, based on normal distribution random sampling, feature extraction is performed on the historical cement matrix images of the binary image training samples to obtain noise vectors. The image feature vector and the noise vector are fused to obtain a fused vector. The fused vector is then upsampled to obtain an initial microscopic image of the cement structure. Feature extraction is performed on the initial microscopic image of the cement structure and the binary image of the binary image training sample to obtain a first feature vector and a second feature vector. The difference between the first feature vector and the second feature vector is calculated to obtain the difference calculation result.

[0011] Optionally, in the method for generating microscopic images of cement structures based on multi-point statistical modification semantics, the difference calculation between the first feature vector and the second feature vector specifically involves: ; in, The second eigenvector distribution and the distribution of the first eigenvector Wassertein distance, For the expected upper bound, For function The conditions that are met For the Lipschitz continuity condition, For function Second eigenvector distribution The expectations below For function First eigenvector distribution The expectations below This is a microscopic view of the initial cement structure.

[0012] Optionally, in the method for generating microscopic images of cement structures based on multi-point statistical modification semantics, the step of calculating the discriminator loss based on the difference calculation results specifically includes: ; The generator loss is calculated based on the difference calculation results, specifically as follows: ; in, For the discriminator loss value, Microscopic image of the initial cement structure First eigenvector Second eigenvector distribution The expectations below This is the structural micrograph corresponding to the fusion vector. The eigenvectors of the structural micrograph In the distribution of fusion vector The expectations below This represents the generator loss value.

[0013] Optionally, the method for generating microscopic images of cement structures based on multi-point statistical modification semantics, wherein the system for generating microscopic images of cement structures based on multi-point statistical modification semantics includes: The data processing module is used to acquire a historical image training set of cement matrix, and to preprocess the historical image training set to obtain a binary image training set. The model training module is used to perform multi-point statistical training on the created target generative adversarial neural network using the binary image training set to obtain a structural microscopic image generation model. The image generation module is used to acquire target image data of the target cement matrix, input the target image data into the structural micro-image generation model, and output the cement structure micro-image of the target cement matrix.

[0014] Furthermore, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a cement structure micro-image generation program based on multi-point statistical modification semantics stored in the memory and executable on the processor. When the cement structure micro-image generation program based on multi-point statistical modification semantics is executed by the processor, it implements the steps of the cement structure micro-image generation method based on multi-point statistical modification semantics as described above.

[0015] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a cement structure micro-image generation program based on multi-point statistical modification semantics, and when the cement structure micro-image generation program based on multi-point statistical modification semantics is executed by a processor, it implements the steps of the cement structure micro-image generation method based on multi-point statistical modification semantics as described above.

[0016] In this invention, a historical image training set of cement matrix is ​​obtained, and the historical image training set is preprocessed to obtain a binary image training set. Based on the binary image training set, a target generative adversarial neural network is trained using multi-point statistical methods to obtain a structural micro-image generation model. Target image data of the target cement matrix is ​​obtained and input into the structural micro-image generation model to output a cement structure micro-image of the target cement matrix. This invention uses a multi-point statistical algorithm to scan images to extract information, achieving automatic annotation of the latent spatial semantics of the generative adversarial neural network, improving the controllability of generating cement micro-structure images, and fusing the extracted information into a noise vector, thereby improving the efficiency and accuracy of generating cement micro-structure images. Attached Figure Description

[0017] Figure 1 This is a flowchart of a preferred embodiment of the method for generating microscopic images of cement structures based on multi-point statistical modification semantics of the present invention; Figure 2 This is a schematic diagram of the overall process of the method for generating microscopic images of cement structures based on multi-point statistical modification semantics according to the present invention; Figure 3 A schematic diagram of the extraction process of the SNESIM algorithm in a preferred embodiment of the present invention; Figure 4 A schematic diagram of a probability information tree table in a preferred embodiment of the present invention; Figure 5 This is a structural diagram of a preferred embodiment of the cement structure micrograph generation system based on multi-point statistical modification semantics of the present invention; Figure 6 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] The preferred embodiment of the present invention describes a method for generating microscopic images of cement structures based on multi-point statistical modification semantics, such as... Figure 1 As shown, the method for generating microscopic images of cement structures based on multi-point statistical modification semantics includes the following steps: Step S10: Obtain a historical image training set of cement matrix, and preprocess the historical image training set to obtain a binary image training set.

[0020] Specifically, addressing the issues of low controllability, efficiency, and accuracy in generating cement microstructure images using existing technologies, this invention proposes a method for generating cement structure microstructure images based on multi-point statistical modification semantics. The specific processing steps of this method are as follows: Figure 2 As shown, a historical image training set of cement matrix is ​​obtained, which includes multiple sets of historical cement matrix images and corresponding historical cement matrix structure microscopic images. Before scanning the historical image training set using the multi-point statistical algorithm SNESIM (a stochastic simulation method based on multi-point geostatistics), the historical image training set needs to be preprocessed, i.e., binarized. The corresponding binarization process is as follows: pixel values ​​are extracted from all historical cement matrix structure microscopic images in the historical image training set to obtain image pixel value extraction results. All image pixel values ​​in the image pixel value extraction results are compared with a preset threshold (e.g., 127). If there is a first image pixel value greater than or equal to the preset threshold, the first image pixel value is set to a first value (e.g., 1), resulting in a first setting result. If there is a second image pixel value less than the preset threshold, the first image pixel value is set to a second value (e.g., 0), resulting in a second setting result. Multiple binary images are obtained based on the first setting result and the second setting result. All the binary images are combined with the corresponding historical cement matrix images to obtain a binary image training set.

[0021] Step S20: Perform multi-point statistical training on the created target generative adversarial neural network based on the binary image training set to obtain a structural microscopic image generation model.

[0022] Specifically, in this embodiment of the invention, the target generative adversarial neural network (GAN) is subjected to multi-point statistical training based on the binary graph training set. This includes two main steps: Step 1 involves scanning the binary graph training set using the multi-point statistical algorithm SNESIM to extract conditional probability information (i.e., a probability information tree table); Step 2 involves fusing the extracted conditional probability information with the noise vector in the latent space and training the model using a Wassertein adversarial neural network (i.e., the target GAN, which includes a generator network and a discriminator network). The Wassertein adversarial neural network is a neural network that incorporates Wasserstein distance as an optimization metric. Wasserstein distance, also known as EMD (Earth Mover's Distance, a metric used to measure the difference between two probability distributions), is used to create the target GAN and input a set of binary graph training samples from the binary graph training set into the target GAN.

[0023] For step one, such as Figure 3 As shown, obtain the image pixel search template (the corresponding shape is as follows). Figure 3 The search template in, among which, The center pixel of the image pixel search template. For the upper condition pixel points of the image pixel search template, For the right condition pixel of the image pixel search template, For image pixel search template, the lower condition pixel point, (The left condition pixel of the image pixel search template) The binary image of the binary image training sample is scanned pixel by pixel according to the image pixel search template to obtain the pixel scan result, and the pixel scan result is extracted to obtain the pixel extraction result (e.g. Figure 3 The extraction results are then used to build and update a binary tree. Specifically, a binary tree is built based on the pixel extraction results, and the root node of the binary tree records the pixel search template in the image. The quantity of 0s and 1s, and according to The value of is used to find the next child node, where a value of 0 indicates the left child node, and a value of 1 indicates the right child node. After finding the child node, according to... The node element values ​​are updated to update the binary tree. This involves updating the binary tree to obtain the target binary tree, calculating the probability count of all pixels in the target binary tree, obtaining the calculation result, and then generating a probability information tree table based on the calculation result. The corresponding structure is as follows: Figure 4As shown, this probability information tree table stores the conditional probability information of the binary image. If the values ​​of the surrounding conditional pixels are known, the probability that the center pixel is 0 given the values ​​of the surrounding conditional pixels can be obtained according to the conditional probability formula. The conditional probability formula is: ; in, The value of the pixel in the above condition. The value of the right condition pixel. For the first Conditional pixels, For the first The value of each condition pixel; the number of center pixels with a value of 0 is obtained according to the counting formula. The corresponding calculation formula is: .

[0024] Subsequently, feature extraction is performed on the historical cement matrix images of the binary image training samples according to the probability information tree table to obtain image feature vectors, and feature extraction is performed on the historical cement matrix images of the binary image training samples according to normal distribution random sampling to obtain noise vectors; the image feature vectors and the noise vectors are fused to obtain fused vectors.

[0025] For step two, the fusion vector obtained in step one is used as input to the generator network in the Wassertein adversarial neural network for model training. During training, the generator network uses upsampling to generate an image from the vector, i.e., an initial microscopic image of the cement structure is obtained based on the fusion vector. Upsampling involves inserting zero values ​​between pixels in each row and column to increase the image size. Then, the generated result (i.e., the initial microscopic image of the cement structure) is input into the discriminator network in the Wassertein adversarial neural network. The discriminator network uses a convolutional neural network to extract features from the initial microscopic image of the cement structure. A self-attention layer is added to the extracted feature map, and then a first feature vector is output after forward computation. The Wassertein distance is calculated between this first feature vector and the second feature vector of the binary image of the binary image training sample. Specifically, features are extracted from both the initial microscopic image of the cement structure and the binary image of the binary image training sample to obtain a first feature vector and a second feature vector. The difference between the first and second feature vectors is calculated, and the formula corresponding to the difference calculation result is as follows: ; in, The second eigenvector distribution and the distribution of the first eigenvector Wassertein distance, For the expected upper bound, For function The conditions that are met For the Lipschitz continuity condition, For function Second eigenvector distribution The expectations below For function First eigenvector distribution The expectations below This is a microscopic view of the initial cement structure.

[0026] Then, based on the difference calculation results, the discriminator loss is calculated to obtain the discriminator loss value. The corresponding formula is: ; in, Microscopic image of the initial cement structure First eigenvector Second eigenvector distribution The expectations below This is the structural micrograph corresponding to the fusion vector. The eigenvectors of the structural micrograph In the distribution of fusion vector The expected value is calculated based on the difference calculation results. The generator loss is then calculated to obtain the generator loss value. The corresponding formula is: ; The parameters of the target generative adversarial neural network are corrected based on the discriminator loss and the generator loss; the next set of binary image training samples is input into the target generative adversarial neural network, and the above process is repeated, which will not be elaborated here; until the training status of the target generative adversarial neural network meets the preset conditions, which can be determined based on the accuracy of the target generative adversarial neural network, which will not be explained in detail here, and finally the trained structural microscopic image generation model is obtained.

[0027] Step S30: Obtain target image data of the target cement matrix, input the target image data into the structural micro-image generation model, and output the cement structure micro-image of the target cement matrix.

[0028] Specifically, in this embodiment of the invention, after obtaining the trained structural microscopic image generation model, XCT images can be generated using the structural microscopic image generation model. The corresponding process is as follows: acquiring target image data of the target cement matrix, inputting the target image data into the structural microscopic image generation model, and outputting a cement structure microscopic image of the target cement matrix. This invention uses a multi-point statistical algorithm to scan the image to extract information, achieving automatic annotation of the latent spatial semantics of the generative adversarial neural network, improving the controllability of generating cement microscopic structure images, and fusing the extracted information into a noise vector, thereby improving the efficiency and accuracy of generating cement microscopic structure images.

[0029] Furthermore, such as Figure 5 As shown, based on the above-mentioned method for generating microscopic images of cement structures based on multi-point statistical modification semantics, the present invention also provides a system for generating microscopic images of cement structures based on multi-point statistical modification semantics, wherein the system for generating microscopic images of cement structures based on multi-point statistical modification semantics includes: Data processing module 51 is used to acquire a historical image training set of cement matrix, and preprocess the historical image training set to obtain a binary image training set; Model training module 52 is used to perform multi-point statistical training on the created target generative adversarial neural network using the binary image training set to obtain a structural microscopic image generation model. The image generation module 53 is used to acquire target image data of the target cement matrix, input the target image data into the structural micro-image generation model, and output the cement structure micro-image of the target cement matrix.

[0030] Furthermore, such as Figure 6 As shown, based on the above-mentioned method for generating microscopic images of cement structures based on multi-point statistical modification semantics, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 6 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0031] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal. Further, the memory 20 may include both internal and external storage units of the terminal. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 may also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a cement structure micro-image generation program 40 based on multi-point statistical modification semantics. This cement structure micro-image generation program 40 based on multi-point statistical modification semantics can be executed by the processor 10, thereby implementing the cement structure micro-image generation method based on multi-point statistical modification semantics in this application.

[0032] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the method for generating microscopic images of cement structures based on multi-point statistical modification semantics.

[0033] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface.

[0034] In one embodiment, when processor 10 executes the cement structure micrograph generation program 40 based on multi-point statistical modification semantics in memory 20, the following steps are performed: A historical image training set of cement matrix is ​​obtained, and the historical image training set is preprocessed to obtain a binary image training set; Based on the binary image training set, the created target generative adversarial neural network is subjected to multi-point statistical training to obtain a structural microscopic image generation model. Acquire target image data of the target cement matrix, input the target image data into the structural micro-image generation model, and output the cement structure micro-image of the target cement matrix.

[0035] The step of obtaining a historical image training set of cement-based materials, and preprocessing the historical image training set to obtain a binary image training set, specifically includes: A training set of historical images of cement matrix is ​​obtained, wherein the training set of historical images includes multiple sets of historical cement matrix images and corresponding historical cement matrix structure micro images; Pixel values ​​are extracted from all historical cement matrix structure micro images in the historical image training set to obtain image pixel value extraction results, and all image pixel values ​​in the image pixel value extraction results are compared with a preset threshold. If there is a first image pixel value that is greater than or equal to the preset threshold, then the first image pixel value is set to a first value to obtain a first setting result; If there is a second image pixel value that is less than the preset threshold, then the first image pixel value is set to the second value to obtain the second setting result; Multiple binary images are obtained based on the first setting result and the second setting result, and all the binary images are combined with the corresponding historical cement matrix images to obtain a binary image training set.

[0036] Specifically, the step of performing multi-point statistical training on the created target generative adversarial neural network based on the binary image training set to obtain a structural microscopic image generation model includes: Create a target generative adversarial neural network, and input a set of binary image training samples from the binary image training set into the target generative adversarial neural network; Multi-point statistical scanning is performed on the binary image of the training sample to obtain pixel extraction results. Based on the pixel extraction results, binary tree processing is performed to obtain a probability information tree table. The probability information tree table is fused into a vector to obtain a fusion vector. An initial cement structure micro-image is obtained based on the fusion vector. The initial cement structure micro-image and the binary image of the binary image training sample are then subjected to a difference calculation to obtain the difference calculation result. The discriminator loss is calculated based on the difference calculation results to obtain the discriminator loss value. The generator loss is calculated based on the difference calculation results to obtain the generator loss value. The parameters of the target generative adversarial neural network are then corrected based on the discriminator loss and the generator loss. The next set of binary image training samples is input into the target generative adversarial neural network until the training status of the target generative adversarial neural network meets the preset conditions, thus obtaining a trained structural microscopic image generation model.

[0037] Specifically, the step of performing multi-point statistical scanning on the binary image of the training samples to obtain pixel extraction results, and then performing binary tree processing based on the pixel extraction results to obtain a probability information tree table, includes: Obtain an image pixel search template, scan the binary image of the binary image training sample pixel by pixel according to the image pixel search template to obtain pixel scanning results, and extract pixels from the pixel scanning results to obtain pixel extraction results; A binary tree is built based on the pixel extraction results. The binary tree is then updated to obtain the target binary tree. The count probability values ​​of all pixels in the target binary tree are calculated to obtain the calculation results. A probability information tree table is then obtained based on the calculation results.

[0038] Specifically, the process of vector fusion of the probability information tree table to obtain a fusion vector, obtaining an initial cement structure micromap based on the fusion vector, and performing difference calculations on the initial cement structure micromap and the binary images of the binary image training samples to obtain difference calculation results includes: Based on the probability information tree table, feature extraction is performed on the historical cement matrix images of the binary image training samples to obtain image feature vectors. Then, based on normal distribution random sampling, feature extraction is performed on the historical cement matrix images of the binary image training samples to obtain noise vectors. The image feature vector and the noise vector are fused to obtain a fused vector. The fused vector is then upsampled to obtain an initial microscopic image of the cement structure. Feature extraction is performed on the initial microscopic image of the cement structure and the binary image of the binary image training sample to obtain a first feature vector and a second feature vector. The difference between the first feature vector and the second feature vector is calculated to obtain the difference calculation result.

[0039] Specifically, the difference calculation between the first feature vector and the second feature vector is performed as follows: ; in, The second eigenvector distribution and the distribution of the first eigenvector Wassertein distance, For the expected upper bound, For function The conditions that are met For the Lipschitz continuity condition, For function Second eigenvector distribution The expectations below For function First eigenvector distribution The expectations below This is a microscopic view of the initial cement structure.

[0040] Specifically, the discriminator loss calculation based on the difference calculation result includes: ; The generator loss is calculated based on the difference calculation results, specifically as follows: ; in, For the discriminator loss value, Microscopic image of the initial cement structure First eigenvector Second eigenvector distribution The expectations below This is the structural micrograph corresponding to the fusion vector. The eigenvectors of the structural micrograph In the distribution of fusion vector The expectations below This represents the generator loss value.

[0041] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a cement structure micro-image generation program based on multi-point statistical modification semantics, and the cement structure micro-image generation program based on multi-point statistical modification semantics implements the steps of the cement structure micro-image generation method based on multi-point statistical modification semantics as described above when executed by a processor.

[0042] In summary, this invention provides a method, system, terminal, and storage medium for generating microscopic images of cement structures based on multi-point statistical semantic modification. The method includes: acquiring a historical image training set of cement substrates; preprocessing the historical image training set to obtain a binary image training set; performing multi-point statistical training on a created target generative adversarial neural network based on the binary image training set to obtain a structural microscopic image generation model; acquiring target image data of the target cement substrate; inputting the target image data into the structural microscopic image generation model; and outputting a cement structural microscopic image of the target cement substrate. This invention uses a multi-point statistical algorithm to scan images to extract information, achieving automatic annotation of the latent spatial semantics of the generative adversarial neural network, improving the controllability of generating cement microscopic structure images, and fusing the extracted information into a noise vector, thereby improving the efficiency and accuracy of generating cement microscopic structure images.

[0043] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0044] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0045] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for generating microscopic images of cement structures based on multi-point statistical modification semantics, characterized in that, The method for generating microscopic images of cement structures based on multi-point statistical modification semantics includes: A historical image training set of cement matrix is ​​obtained, and the historical image training set is preprocessed to obtain a binary image training set; Based on the binary image training set, the created target generative adversarial neural network is subjected to multi-point statistical training to obtain a structural microscopic image generation model. The step of performing multi-point statistical training on the created target generative adversarial neural network based on the binary image training set to obtain a structural microscopic image generation model specifically includes: Create a target generative adversarial neural network, and input a set of binary image training samples from the binary image training set into the target generative adversarial neural network; Multi-point statistical scanning is performed on the binary image of the training sample to obtain pixel extraction results. Based on the pixel extraction results, binary tree processing is performed to obtain a probability information tree table. The probability information tree table is fused into a vector to obtain a fusion vector. An initial cement structure micro-image is obtained based on the fusion vector. The initial cement structure micro-image and the binary image of the binary image training sample are then subjected to a difference calculation to obtain the difference calculation result. The discriminator loss is calculated based on the difference calculation results to obtain the discriminator loss value. The generator loss is calculated based on the difference calculation results to obtain the generator loss value. The parameters of the target generative adversarial neural network are then corrected based on the discriminator loss and the generator loss. The next set of binary image training samples is input into the target generative adversarial neural network until the training status of the target generative adversarial neural network meets the preset conditions, and a trained structural microscopic image generation model is obtained. Acquire target image data of the target cement matrix, input the target image data into the structural micro-image generation model, and output the cement structure micro-image of the target cement matrix.

2. The method for generating microscopic images of cement structures based on multi-point statistical modification semantics according to claim 1, characterized in that, The process of obtaining a historical image training set of cement-based materials, and preprocessing the historical image training set to obtain a binary image training set, specifically includes: A training set of historical images of cement matrix is ​​obtained, wherein the training set of historical images includes multiple sets of historical cement matrix images and corresponding historical cement matrix structure micro images; Pixel values ​​are extracted from all historical cement matrix structure micro images in the historical image training set to obtain image pixel value extraction results, and all image pixel values ​​in the image pixel value extraction results are compared with a preset threshold. If there is a first image pixel value that is greater than or equal to the preset threshold, then the first image pixel value is set to a first value to obtain a first setting result; If there is a second image pixel value that is less than the preset threshold, then the first image pixel value is set to the second value to obtain the second setting result; Multiple binary images are obtained based on the first setting result and the second setting result, and all the binary images are combined with the corresponding historical cement matrix images to obtain a binary image training set.

3. The method for generating microscopic images of cement structures based on multi-point statistical modification semantics according to claim 1, characterized in that, The step of performing multi-point statistical scanning on the binary image of the training samples to obtain pixel extraction results, and then performing binary tree processing based on the pixel extraction results to obtain a probability information tree table, specifically includes: Obtain an image pixel search template, scan the binary image of the binary image training sample pixel by pixel according to the image pixel search template to obtain pixel scanning results, and extract pixels from the pixel scanning results to obtain pixel extraction results; A binary tree is built based on the pixel extraction results. The binary tree is then updated to obtain the target binary tree. The count probability values ​​of all pixels in the target binary tree are calculated to obtain the calculation results. A probability information tree table is then obtained based on the calculation results.

4. The method for generating microscopic images of cement structures based on multi-point statistical modification semantics according to claim 1, characterized in that, The process involves vector fusion of the probability information tree table to obtain a fusion vector, obtaining an initial cement structure micromap based on the fusion vector, and performing difference calculations on the initial cement structure micromap and the binary images of the binary image training samples to obtain difference calculation results. Specifically, this includes: Based on the probability information tree table, feature extraction is performed on the historical cement matrix images of the binary image training samples to obtain image feature vectors. Then, based on normal distribution random sampling, feature extraction is performed on the historical cement matrix images of the binary image training samples to obtain noise vectors. The image feature vector and the noise vector are fused to obtain a fused vector. The fused vector is then upsampled to obtain an initial microscopic image of the cement structure. Feature extraction is performed on the initial microscopic image of the cement structure and the binary image of the binary image training sample to obtain a first feature vector and a second feature vector. The difference between the first feature vector and the second feature vector is calculated to obtain the difference calculation result.

5. The method for generating microscopic images of cement structures based on multi-point statistical modification semantics according to claim 4, characterized in that, The difference calculation between the first feature vector and the second feature vector specifically involves: ； in, The second eigenvector distribution and the distribution of the first eigenvector Wassertein distance, For the upper bound of expectation, For function The conditions that are met For the Lipschitz continuity condition. For function Second eigenvector distribution The expectations below For function First eigenvector distribution The expectations below This is a microscopic view of the initial cement structure.

6. The method for generating microscopic images of cement structures based on multi-point statistical modification semantics according to claim 1, characterized in that, The discriminator loss calculation based on the difference calculation result is specifically as follows: ； The generator loss is calculated based on the difference calculation results, specifically as follows: ； in, For the discriminator loss value, Microscopic image of the initial cement structure First eigenvector Second eigenvector distribution The expectations below This is the structural micrograph corresponding to the fusion vector. The eigenvectors of the structural micrograph In the distribution of fusion vector The expectations below This represents the generator loss value.

7. A system for generating microscopic images of cement structures based on multi-point statistical modification semantics, characterized in that, The cement structure micro-image generation system based on multi-point statistical modification semantics is used to implement the cement structure micro-image generation method based on multi-point statistical modification semantics as described in any one of claims 1-6. The cement structure micro-image generation system based on multi-point statistical modification semantics includes: The data processing module is used to acquire a historical image training set of cement matrix, and to preprocess the historical image training set to obtain a binary image training set. The model training module is used to perform multi-point statistical training on the created target generative adversarial neural network using the binary image training set to obtain a structural microscopic image generation model. The image generation module is used to acquire target image data of the target cement matrix, input the target image data into the structural micro-image generation model, and output the cement structure micro-image of the target cement matrix.

8. A terminal, characterized in that, The terminal includes a memory, a processor, and a program stored in the memory and executable on the processor. When the program is executed by the processor, it implements the steps of the method for generating microscopic images of cement structures based on multi-point statistical modification semantics as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which stores a cement structure micro-image generation program based on multi-point statistical modification semantics. When the cement structure micro-image generation program based on multi-point statistical modification semantics is executed by a processor, it implements the steps of the cement structure micro-image generation method based on any one of claims 1-6.

Images