A method and apparatus for modal conversion of cement hydration image data

Abstract

The application provides a cement hydration image data modal conversion method and device, which can be used in the field of artificial intelligence technology, and the method comprises the following steps: scanning a hardened cement paste sample, constructing a sample data set according to CT sample data and BSE sample data obtained through scanning; training a preset diffusion model through the training data set, constructing a modal conversion model, performing modal conversion on CT data to be converted, and generating BSE data, which realizes the modal conversion of X-CT images of cement hydration microstructure to BSE images, can effectively combine the time-space imaging advantages of X-CT and the high resolution and contrast advantages of BSE, thereby greatly improving the spatial resolution of the hydration image, ensuring the evolution imaging of the full time scale, fully reflecting the whole process of cement hydration, and reducing the cost; at the same time, the hydration products with similar densities can be effectively distinguished, the details in the hydration process can be accurately captured, and four-dimensional imaging of the cement hydration process can be realized.

Description

Technical Field

[0001] This invention relates to the field of computer technology, particularly to the field of artificial intelligence technology, and especially to a method and apparatus for modal conversion of cement hydration image data. Background Technology

[0002] The key to studying the performance of cement concrete lies in its microstructure. Microstructure originates from the development of cement hydration, and direct characterization of cement hydration, especially non-destructive dynamic testing, has long been a focus of the industry, and also a technological bottleneck.

[0003] Among related technologies, focused ion beam electron microscopy (FIB-SEM) backscattering technology can obtain high-resolution microscopic images of the cement hydration process and accurately characterize the microstructure of cement. However, FIB-SEM is a destructive testing technique and is costly. The sample is polished and cut under the action of the ion beam, which makes it impossible to retain the time information of the hydration process. At the same time, the cutting range of the sample is small, resulting in the obtained data results lacking broad spatial representativeness and making it difficult to fully reflect the entire process of cement hydration.

[0004] Another related technique uses super-resolution methods to interpolate low-resolution X-ray computed tomography (X-CT) images, generating new pixels to improve the resolution of the X-CT images. However, X-CT images themselves have limitations, especially between substances with similar densities (e.g., high-density CSH versus calcium hydroxide and ettringite). Due to the physical limitations of X-ray attenuation, these substances are often difficult to distinguish effectively. Even with super-resolution algorithms to improve image resolution, it is still difficult to overcome this difficulty in distinguishing between substances, thus failing to accurately capture details in the hydration process. Traditional interpolation and reconstruction algorithms can only improve the pixel resolution of images to a certain extent, and cannot truly improve the spatial resolution of images, failing to accurately represent the details of hydration. Summary of the Invention

[0005] One objective of this invention is to provide a modal conversion method for cement hydration image data, realizing the modal conversion of X-CT images of cement hydration microstructures to BSE images. This method effectively combines the temporal-spatial imaging advantages of X-CT with the high resolution and contrast advantages of BSE, thereby significantly improving the spatial resolution of hydration images, ensuring evolutionary imaging across the entire timescale, comprehensively reflecting the entire cement hydration process, and reducing costs. Simultaneously, it effectively distinguishes hydration products with similar densities, accurately captures details in the hydration process, and achieves four-dimensional imaging of the cement hydration process. Another objective of this invention is to provide a modal conversion device for cement hydration image data. A further objective of this invention is to provide a computer-readable medium. A final objective of this invention is to provide a computer device.

[0006] To achieve the above objectives, this invention discloses a modal conversion method for cement hydration image data, comprising: Prepare hardened cement slurry samples; By scanning hardened cement slurry samples, three-dimensional imaging sample data and electron microscope scanning sample data of cement hydration were obtained; A sample dataset is constructed based on 3D imaging sample data and electron microscope scanning sample data. The sample dataset includes the training dataset. By training the training dataset, a pre-defined diffusion model is trained to construct a modality transformation model; The modal transformation model is used to perform modal transformation on the 3D imaging data to be transformed, generating electron microscope scanning data.

[0007] Preferably, the preparation of a hardened cement slurry sample includes: Prepare hardened cement slurry at different ages; The hardened cement slurry was fixed and sliced ​​to obtain hardened cement slurry samples.

[0008] Preferably, three-dimensional imaging sample data and electron microscopy scanning sample data of cement hydration are obtained by scanning hardened cement slurry samples, including: The hardened cement slurry sample was scanned and imaged using computed tomography (CT) technology to generate three-dimensional imaging sample data. The hardened cement slurry samples after computed tomography (CT) scan were carbonized, and the carbonized hardened cement slurry samples were scanned using electron microscopy backscattering technology to generate electron microscopy scan sample data.

[0009] Preferably, a sample dataset is constructed based on three-dimensional imaging sample data and electron microscopy scanning sample data, including: Based on the 3D imaging sample data and the electron microscope scanning sample data, a sample mapping relationship is established, which includes the 3D imaging sample data and the corresponding electron microscope scanning sample data. Using an image registration algorithm, spatial registration and distortion correction are performed on 3D imaging sample data and corresponding electron microscope scan sample data to generate paired sample data; Construct a sample dataset based on multiple pairs of paired sample data; The sample dataset is divided into a training dataset and a validation dataset.

[0010] Preferably, before training the pre-defined diffusion model using the training dataset to construct the modality transformation model, the method further includes: The training dataset is augmented to generate an augmented training dataset. The augmentation process includes rotation, region cutting, random flipping, random grayscale adjustment, and random brightness adjustment.

[0011] Preferably, a modality transformation model is constructed by training a pre-defined diffusion model using a training dataset, including: Using electron microscope scanned sample data as a monitoring signal, the electron microscope scanned sample data is forward diffused according to a preset noise signal at a preset time step to generate noise image data. According to a preset time step, the predicted noise signal is generated in reverse based on the noise image data. Using the three-dimensional imaging sample data corresponding to the electron microscope scanning sample data as constraints, the conditions are generated based on the predicted noise signal and noise image data to obtain the restored electron microscope scanning data. By using a loss function, backpropagation is performed based on the loss result between the restored electron microscope scan data and the electron microscope scan sample data to iteratively update the model parameters and construct a mode transformation model.

[0012] Preferably, the method further includes: The modality conversion model was tested using the constructed test dataset, and electron microscope scanning test data was obtained. The volume fraction was compared between electron microscope scanning test data and actual measurement data to generate model validation results.

[0013] The present invention also discloses a modal conversion device for cement hydration image data, comprising: Cement slurry sample preparation unit, used to prepare hardened cement slurry samples; The sample data acquisition unit is used to acquire three-dimensional imaging sample data and electron microscope scanning sample data of cement hydration by scanning hardened cement slurry samples; The sample dataset construction unit is used to construct a sample dataset based on 3D imaging sample data and electron microscope scanning sample data. The sample dataset includes the training dataset. The model training unit is used to train a pre-defined diffusion model using a training dataset to build a modality transformation model. The modality conversion unit is used to perform modality conversion on the 3D imaging data to be converted through the modality conversion model, generating electron microscope scanning data.

[0014] The present invention also discloses a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0015] The present invention also discloses a computer device, including a memory and a processor, wherein the memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions, wherein the processor executes the program to implement the method described above.

[0016] The present invention also discloses a computer program product, including a computer program / instruction, which, when executed by a processor, implements the method described above.

[0017] This invention prepares hardened cement slurry samples; by scanning the hardened cement slurry samples, it obtains three-dimensional imaging sample data and electron microscope scanning sample data of cement hydration; based on the three-dimensional imaging sample data and electron microscope scanning sample data, it constructs a sample dataset, which includes a training dataset; through the training dataset, it trains a preset diffusion model to construct a modal transformation model; through the modal transformation model, it performs modal transformation on the three-dimensional imaging data to be transformed, generating electron microscope scanning data, realizing the modal transformation of X-CT images of cement hydration microstructure to BSE images. This effectively combines the temporal-spatial imaging advantages of X-CT with the high resolution and contrast advantages of BSE, thereby significantly improving the spatial resolution of hydration images, ensuring evolution imaging across the entire time scale, comprehensively reflecting the entire process of cement hydration, and reducing costs; at the same time, it effectively distinguishes hydration products with similar densities, accurately captures details in the hydration process, and realizes four-dimensional imaging of the cement hydration process. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1A flowchart of a modal conversion method for cement hydration image data provided in an embodiment of the present invention; Figure 2 A flowchart illustrating another modal conversion method for cement hydration image data provided in this embodiment of the invention; Figure 3 This is a schematic diagram of the structure of a modal conversion device for cement hydration image data provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] It should be noted that the modal conversion method and apparatus for cement hydration image data disclosed in this application can be used in the field of artificial intelligence technology, or in any field other than artificial intelligence technology. The application field of the modal conversion method and apparatus for cement hydration image data disclosed in this application is not limited.

[0022] To facilitate understanding of the technical solutions provided in this application, the relevant content of the technical solutions will be explained below. Electron backscattering microscopy (BSE) can effectively characterize the microstructure of cement during the hydration process, acquiring high-contrast, high-resolution images and identifying features such as unhydrated minerals, high-density hydrated calcium silicate (CSH), low-density CSH, calcium hydroxide, ettringite, and pores in cement. However, BSE requires complex sample pretreatment, which inevitably interrupts the hydration process, resulting in a loss of time-scale evolution information. Furthermore, BSE is only applied to the sample surface and cannot acquire internal spatial feature information. On the other hand, X-ray computed tomography (X-CT) can reconstruct the three-dimensional structure of cement during the hydration process, achieving dynamic characterization of the hydration structure. Compared to BSE, X-CT can perform in-situ non-destructive testing, effectively characterizing the time-varying evolution of cement hydration structures. However, the imaging scale of X-CT is limited by the physical constraints of sample size and the X-ray source, which is difficult to overcome. This results in limitations on resolution due to sample size, as well as the inherent resolution limitations of X-CT technology itself. Nevertheless, due to the advantages of non-destructive detection and the low requirements for sample processing, X-CT remains an important technique for characterizing the four-dimensional (time + three-dimensional space) evolution of hydration structures.

[0023] In recent years, significant progress has been made in the application of artificial intelligence methods, especially deep learning technology, in the field of computer vision. By leveraging generative models and using X-CT data as a foundation, combined with homologous heterogeneous BSE image data as supervision, modality conversion from X-CT data to BSE images can be achieved. This combines the four-dimensional advantages of X-CT with the high contrast and high resolution advantages of BSE to decode and reconstruct the spatial evolution features of the cement hydration process.

[0024] This invention achieves four-dimensional imaging of cement hydration at the nano-micro-mesh level based on a diffusion-generative deep learning model. Specifically, firstly, the invention acquires X-CT and BSE images (these two types of images exhibit homologous heterogeneity) during the cement hydration process, and establishes spatial registration relationships between the two types of images through optimized experimental methods, constructing a training dataset. Secondly, a CT-BSE modal conversion generative model based on the diffusion-generative model is constructed. Finally, the model is trained using the established dataset to achieve modal conversion from X-CT images of the cement hydration microstructure to BSE images. This method effectively combines the temporal-spatial imaging advantages of X-CT with the high resolution and contrast advantages of BSE, thereby significantly improving the spatial resolution of hydration images (by more than 10 times), ensuring temporal evolution imaging, and effectively distinguishing hydration products with similar densities, thus achieving four-dimensional imaging of the cement hydration process.

[0025] The following uses a modal conversion device for cement hydration image data as an example to illustrate the implementation process of the modal conversion method for cement hydration image data provided in this embodiment of the invention. It is understood that the execution entity of the modal conversion method for cement hydration image data provided in this embodiment of the invention includes, but is not limited to, a modal conversion device for cement hydration image data.

[0026] Figure 1 A flowchart of a modal conversion method for cement hydration image data provided in an embodiment of the present invention is shown below. Figure 1 As shown, the method includes: Step 101: Prepare a hardened cement slurry sample.

[0027] In this embodiment of the invention, multiple sets of hardened cement paste samples at different ages were prepared; the prepared hardened cement paste samples were fixed with epoxy resin, cut into slices, and the surfaces were ground and polished to ensure that the surface of the hardened cement paste samples was flat and suitable for imaging.

[0028] Step 102: Obtain CT sample data and BSE sample data of cement hydration by scanning the hardened cement slurry sample.

[0029] In this embodiment of the invention, slices of the same hardened cement slurry sample are stacked and X-CT scanned to obtain three-dimensional imaging data of cement hydration, which is recorded as CT sample data; slices of the same hardened cement slurry sample are carbonized and BSE scanned to obtain BSE images of cement hydration, which is recorded as BSE sample data.

[0030] Step 103: Construct a sample dataset based on CT sample data and BSE sample data.

[0031] In this embodiment of the invention, the sample dataset includes a training dataset and a validation dataset.

[0032] In this embodiment of the invention, spatial registration and distortion correction are performed on the acquired CT sample data and BSE sample data to ensure accurate correspondence between the registered BSE data and X-CT data, thereby constructing the sample dataset required for modality conversion; the sample dataset is divided into training dataset and validation dataset according to a preset division ratio.

[0033] Step 104: Train the preset diffusion model using the training dataset to construct a modality transformation model.

[0034] In this embodiment of the invention, a deep learning diffusion model is trained using a training dataset, and hyperparameters are adjusted using a validation dataset to construct a CT-BSE modal conversion model, which serves as the core network for cement hydration image conversion. This modal conversion model is a deep learning model capable of effectively identifying super-resolution features of hydration images.

[0035] Furthermore, the finally trained modal transformation model is used to perform modal transformation on the cement hydration images in the constructed test set, converting them from X-CT images to BSE images. The specific process of constructing the test set is as follows: cement slurry is prepared, and the hydration evolution process of the cement slurry is characterized using X-CT technology. This dataset is used as the test set for subsequent analysis. Thermogravimetric analysis and isothermal calorimetry are used to verify the results of the test set, ensuring the accuracy and reliability of the generated BSE images.

[0036] Step 105: Using the modal transformation model, perform modal transformation on the CT data to be transformed to generate BSE data.

[0037] Specifically, the CT data to be converted is input into the modal conversion model for BSE data modal conversion, and BSE data is output.

[0038] In the technical solution provided by this invention, a hardened cement slurry sample is prepared; the hardened cement slurry sample is scanned to obtain three-dimensional imaging sample data and electron microscope scanning sample data of cement hydration; a sample dataset is constructed based on the three-dimensional imaging sample data and electron microscope scanning sample data, the sample dataset including a training dataset; a preset diffusion model is trained using the training dataset to construct a modal conversion model; the modal conversion model is used to perform modal conversion on the three-dimensional imaging data to be converted, generating electron microscope scanning data, thus realizing the modal conversion of X-CT images of cement hydration microstructure to BSE images. This effectively combines the temporal-spatial imaging advantages of X-CT with the high resolution and contrast advantages of BSE, thereby significantly improving the spatial resolution of hydration images, ensuring evolution imaging across the entire time scale, comprehensively reflecting the entire process of cement hydration, and reducing costs; at the same time, it effectively distinguishes hydration products with similar densities, accurately captures details in the hydration process, and realizes four-dimensional imaging of the cement hydration process.

[0039] Figure 2 A flowchart of another modal conversion method for cement hydration image data provided in an embodiment of the present invention is shown below. Figure 2 As shown, the method includes: Step 201: Prepare hardened cement slurry at different ages.

[0040] It is worth noting that the settings for different age groups can be configured according to actual needs, and this embodiment of the invention does not limit this.

[0041] As an alternative, a cement paste with a water-cement ratio of 0.5 was prepared by mixing ordinary silicate cement with water. After stirring for 1 minute, the paste was poured into a cylindrical mold with a diameter of 10 mm and a height of 100 mm to prepare 40 sets of cylindrical hardened cement paste samples with a diameter of 10 mm and a height of 100 mm. These samples were then treated at different ages (1 day, 3 days, 7 days, and 28 days). For each age, 10 sets of samples were selected and immersed in isopropanol solution to stop the hydration process, thus obtaining cement samples at ages of 1 day (d), 3 days, 7 days, and 28 days.

[0042] As an alternative, a cement paste with a water-cement ratio of 0.5 was prepared by mixing ordinary silicate cement with water. After stirring for 1 minute, the paste was poured into a cylindrical mold with a diameter of 10 mm and a height of 100 mm to prepare a total of 12 samples. The samples were divided into 4 groups of 3 samples each. After 1 day, the samples were removed from the mold and placed in a curing room with a humidity of 95% for curing. For the 4 groups of samples, they were immersed in isopropanol solution at 1 day, 3 days, 7 days and 28 days to terminate their hydration.

[0043] Step 202: Fix and slice the hardened cement slurry to obtain a hardened cement slurry sample.

[0044] As an alternative, a cuboid mold with a length and width of 12 mm and a height of 100 mm is selected. The hardened cement slurry is placed in the center of the mold, and epoxy resin is injected into it to cover the sample, reshaping it into a 12×12×100 mm cuboid composite material. 50 slices are cut into 2 mm thin slices per sample. The sample is then ground and polished to the standard for backscattered electron microscopy to obtain the hardened cement slurry sample.

[0045] Step 203: Use X-CT technology to scan and image the hardened cement slurry sample to generate CT sample data.

[0046] In this embodiment of the invention, the sliced ​​hardened cement slurry samples are stacked along the Z-axis in their original order and scanned using a micron X-CT scanning device. The spatial reconstruction of the images is then performed to obtain the original volume data, namely, the three-dimensional imaging sample data, denoted as CT sample data.

[0047] As an alternative, the X-ray source tube voltage is 60 kV, the tube current is 50 μA, the voxel resolution of the X-CT scan is 4 μm³, the frame rate is 1080 FPS, the sliced ​​hardened cement slurry samples are stacked along the Z-axis in the original order, and volume data at 1d, 3d, 7d, and 28d are obtained by scanning. The final CT sample data has a size of 2500 pixels × 2500 pixels × 2500 pixels, where the height in the Z-axis direction is obtained by stitching together the longitudinal data.

[0048] Step 204: Carbon spraying is performed on the hardened cement slurry sample after X-CT scanning, and the hardened cement slurry sample after carbon spraying is scanned by electron microscopy backscattering technology to generate BSE sample data.

[0049] In this embodiment of the invention, each slice of the hardened cement slurry sample after X-CT scanning is carbon-sprayed, placed in a scanning electron microscope, and subjected to backscatter scanning. By scanning point by point, a planar image of the entire hardened cement slurry sample is obtained, i.e., BSE sample data. Specifically, a complete planar image of the hardened cement slurry sample slice is obtained by sweep stitching.

[0050] As an optional approach, the BSE scanning parameters are set to a resolution of 40 nm / pixel. Alternatively, each slice of the hardened cement slurry sample is scanned at a resolution of 40 nm, acquiring BSE sample data for 1 day, 3 days, 7 days, and 28 days, specifically 50 images per sample. Simultaneously, energy dispersive spectroscopy (EDS) is used to obtain phase composition information at different locations in the images to help distinguish different mineral phases within the cement sample.

[0051] Step 205: Establish a sample mapping relationship based on CT sample data and BSE sample data.

[0052] In this embodiment of the invention, the sample mapping relationship includes CT sample data and the corresponding BSE sample data.

[0053] In this embodiment of the invention, planes of different samples are distinguished based on the stacked gaps reconstructed from X-CT images. These planes correspond one-to-one with the planes of BSE images. Since CT sample data and BSE sample data are spatially homogeneous but heterogeneous, there is a natural mapping relationship between them, expressed as:

[0054] in, This represents the nonlinear relationship between resolution and imaging mode from BSE sample data to CT sample data, where η1 represents statistical noise, slice error, and geometric distortion in BSE sample data imaging.

[0055] Step 206: Using an image registration algorithm, spatial registration and distortion correction are performed on the CT sample data and the corresponding BSE sample data to generate paired sample data.

[0056] In this embodiment of the invention, geometric distortion is corrected first, particularly distortion caused by the cone-beam effect during X-CT scanning. Using an image registration algorithm, combined with the degrees of freedom of XY-plane magnification and rotation in the XYZ directions, a feature point-based scale-invariant feature transform (SIFT) algorithm is used to extract feature points from the X-CT and BSE images. These feature points include, but are not limited to, sample edges, unhydrated particles, and large pores. These feature points are then used to estimate the projection transformation matrix or a three-dimensional rigid / affine transformation matrix, thereby correcting the geometric distortion. When a projection transformation is selected, the registration process is described by a 3×3 homography matrix H estimated from the feature points extracted by the SIFT algorithm, which satisfies the following in two-dimensional homogeneous coordinates: in, These are the coordinates of feature points in the CT sample data; These are the coordinates of feature points in the BSE sample data; H is the homography matrix.

[0057] By minimizing the feature point projection error and correcting the cone-beam distortion, It is eliminated or significantly reduced.

[0058] If rotation and scaling in the XYZ directions are considered simultaneously in three-dimensional space, a rigid or affine transformation matrix T can be used to satisfy the following in three-dimensional homogeneous coordinates: Where T includes rotation, translation, and scaling factors that are isotropic or anisotropic; These are the coordinates of feature points in the CT sample data; These are the coordinates of feature points in the BSE sample data.

[0059] It is worth noting that the rigid or affine transformation matrix T can be estimated based on the feature points extracted by the SIFT algorithm, or it can be obtained by expanding the homography matrix H by filling the Z-dimensional dimension.

[0060] After this transformation, the original mapping relationship becomes: This allows for precise alignment of X-CT and BSE images in the same coordinate system.

[0061] In this embodiment of the invention, since the data scanning pixels are large, in order to improve the efficiency of subsequent model training and save computing power, the CT sample data and BSE are cropped into multiple sub-maps of the same size to ensure that each sub-map corresponds one-to-one, thus obtaining multiple pairs of paired sample data.

[0062] As an alternative, the 2500-pixel × 2500-pixel × 2500-pixel CT sample data will be cropped into multiple 128-pixel × 128-pixel sub-patterns. The BSE sample data will also be cropped into 1280-pixel × 1280-pixel sub-patterns in the same way, ensuring that each sub-pattern corresponds one-to-one, resulting in multiple pairs of paired sample data.

[0063] Step 207: Construct a sample dataset based on multiple pairs of paired sample data.

[0064] Specifically, the set of multiple paired sample data is defined as the sample dataset.

[0065] Step 208: Divide the sample dataset into a training dataset and a validation dataset.

[0066] In this embodiment of the invention, paired sample data for a specified age group is determined as the training dataset, and paired sample data for other age groups is determined as the validation dataset. It is worth noting that the ratio of paired sample data between the training dataset and the validation dataset is 3:1.

[0067] As an alternative, paired sample data with ages of 1d, 3d, and 7d are used as the training dataset, and paired sample data with an age of 28d are used as the validation dataset.

[0068] The following specific example illustrates the process of constructing the training and validation datasets: First, the docking area of ​​the CT stacked data was extracted, and the 2500 pixel × 2500 pixel × 25000 pixel images were optimized into 50 sets of 2500 pixel × 2500 pixel × 10 pixel images, corresponding to 50 BSE images.

[0069] Then, using an image registration algorithm, the CT images and BSE images are spatially registered. From each of the 50 image sets, one corresponding BSE image is extracted as an X-CT image to obtain a one-to-one correspondence between the CT and BSE images, establishing a mapping: CT images are cropped to 128 pixels × 128 pixels, and BSE images are cropped to 1280 pixels × 1280 pixels. Each sample constitutes 300 pairs / layer × 50 layers = 15,000 pairs of CT-BSE paired sample data. Therefore, 15,000 × 12 = 180,000 pairs of paired sample data can be obtained to form a sample dataset. Paired sample data of 1d, 3d, and 7d are used to construct the training dataset, and paired sample data of 28d are used to construct the validation dataset, forming a 3:1 data split.

[0070] Step 209: Perform data augmentation on the training dataset to generate the augmented training dataset.

[0071] In this embodiment of the invention, data augmentation processing includes, but is not limited to, rotation, region segmentation, random flipping, random grayscale adjustment, and random brightness adjustment. Specifically, rotation is random, with a rotation range between 0-180°; region segmentation is random, with a segmentation range of 0.4-1 times the original size; random flipping includes flipping the image top, bottom, left, and right; random grayscale adjustment ranges from 0.8-1.2 times the original grayscale; and random brightness adjustment ranges from 0.8-1.2 times the original brightness.

[0072] In this embodiment of the invention, the augmented training dataset contains a large amount of paired sample data, which improves the features of the dataset. In the subsequent model training process, the augmented training dataset is used for training. By optimizing the generative model, it can effectively generate corresponding BSE images from CT images, and the final modality conversion model is obtained through training, which helps improve the generalization ability of the model training.

[0073] Step 210: Using BSE sample data as a supervision signal, perform forward diffusion on the BSE sample data according to a preset time step and a preset noise signal to generate noise image data.

[0074] The diffusion generation model of this invention is used to generate corresponding BSE images from CT images, and generates BSE images under constraints based on CT image data, rather than unconditionally. The BSE images are used as supervision signals to train the model to generate the transformation from CT images to BSE images. The model design incorporates diffusion processes, generation processes, and CT image conditional generation processes.

[0075] In this embodiment of the invention, the forward diffusion process involves gradually adding noise to the image until the image becomes pure noise. If the input BSE sample data is... Its diffusion trajectory can be represented as:

[0076] For time step t, the current BSE sample data Based on the BSE sample data from the previous time step (t-1) and noise level β t The update will be performed in the following form:

[0077] in, β represents the current BSE sample data at time step t; t Noise level; This represents the BSE sample data from the previous time step (t-1); t The noise is noise, and the noise follows a standard normal distribution. t ~N(0,1).

[0078] During this process, the image gradually becomes blurred until it is eventually completely transformed into noise. The forward diffusion process follows a Markov chain, and its joint conditional probability can be expressed as:

[0079] in, For joint conditional probabilities, To generate the conditional probability distribution of BSE sample data for the current time step from the BSE sample data of the previous time step; β represents the current BSE sample data at time step t; t Noise level; I represents the BSE sample data from the previous time step (t-1), and I is the identity matrix.

[0080] This process ensures that noise is gradually added to the image, eventually bringing the image close to a state of completely random noise, resulting in noisy image data.

[0081] Step 211: According to the preset time step, reverse generate the noise signal based on the noise image data to obtain the predicted noise signal.

[0082] In this embodiment of the invention, the essence of the reverse generation process is that, given... In the case of predicting the noise generated from step t-1 to step t, that is: By removing the predicted noise, the BSE image from the previous time step is recovered.

[0083] It's worth noting that this process can be viewed as learning a mapping from noise to a data structure. However, without any constraints, this process becomes completely random, randomly generating BSE images.

[0084] Step 212: Using the CT sample data corresponding to the BSE sample data as constraints, generate conditions based on the predicted noise signal and noise image data to obtain the restored BSE data.

[0085] In this embodiment of the invention, in order to generate BSE sample data from CT sample data, it is necessary to process the CT sample data x CT As a conditional input, by incorporating CT sample data as a condition, the generation process is no longer completely random, but rather generates BSE sample data based on the features of the CT sample data. The conditional generation process can be represented as:

[0086] Through x t BSE With x CT The concatenated input model transforms the generation process from completely random noise input to a process with partial initial condition constraints, limiting the model to consider the CT sample data x at each step of the backdiffusion. CT The information in the sample is used to effectively generate BSE sample data from CT sample data, resulting in restored BSE data.

[0087] Step 213: Using the loss function, backpropagation is performed based on the loss result between the restored BSE data and the BSE sample data to iteratively update the model parameters and construct the modality transformation model.

[0088] In this embodiment of the invention, the dataset is trained for 500 epochs, the loss is calculated on the restored BSE data and BSE sample data, the loss curves of the model discriminator and generator are output using a visualization tool (TensorBoard), the optimal model parameters of each training round are retained and stored as a .pth file based on the loss of the validation dataset, and the optimal model is selected as the modality transformation model based on the final training results.

[0089] Furthermore, by adjusting the hyperparameters of the modality conversion model using a validation dataset, a modality conversion model that can effectively identify super-resolution features of hydration images was finally trained.

[0090] Furthermore, the modality transformation model is tested using the constructed test dataset to obtain BSE test data. Volume fractions are compared between the BSE test data and actual measurement data to generate model validation results. Specifically, the test set is input into the surrogate model for prediction, BSE images are acquired image by image, and reconstructed into a four-dimensional BSE image. The volume fractions and volumes of unhydrated particles, calcium hydroxide (CH), and ettringite (Aft) over time are calculated. The volume fraction is calculated based on the proportion of each substance's pixels to the total number of pixels; the volume is calculated based on the product of the number of pixels of each substance and a voxel, where a voxel is the size of the smallest unit volume.

[0091] The process of constructing the test set is as follows: cement particles are mixed with water in a certain proportion to prepare three samples. The first sample is used to obtain X-CT scan data at the initial moment of the hydration process and at multiple time points. As an optional approach, the time points include 4 hours, 8 hours, 12 hours, 1 day, 2 days, 3 days, 7 days, 28 days, and 90 days, forming four-dimensional data (time dimension + X-CT three-dimensional data). The second sample is used to obtain isothermal calorimetric data to record the heat release during the hydration process. The third sample is subjected to thermogravimetric analysis to measure the change in the mass fraction of water of crystallization. The dynamic evolution process of cement hydration obtained by X-CT is used as the test set, while the latter two (isothermal calorimetric data and thermogravimetric analysis data) are used as indirect verification methods for the model's accuracy.

[0092] It is worth noting that the mixing ratio of cement particles and water can be set according to actual needs, and this embodiment of the invention does not limit this. As an optional solution, the water-cement ratio between cement particles and water is 0.5.

[0093] During the testing process: For unhydrated particles, calcium hydroxide (CH), and ettringite (Aft), the rate of change was calculated separately as an indicator to verify the reliability of the model: After segmentation, the BSE image output by the modality transformation model is used to obtain the number of voxels, and the volumes of unhydrated particles (UH), calcium hydroxide (CH), and ettringite (Aft) in the model output results within one day are obtained. These volumes are then divided by the total volume to obtain the volume fraction of the three components, i.e.: V UH (t), V CH (t) and V Aft (t). The number of voxels can be obtained using any image processing software.

[0094] Calculate its slope:

[0095] Among them, S UH (t), SCH (t) and S Aft (t) represents the slope of the volume fraction change of unhydrated particles, calcium hydroxide, and ettringite within 1 day; t is the hydration age, t≤24h; ΔV UH (t), △V CH (t) and △V Aft (t) represents the difference in volume fraction of unhydrated particles, calcium hydroxide, and ettringite at age t compared to the previous age.

[0096] By comparing the slope of the change in the volume fraction of unhydrated particles in the BSE test data within one day with the corresponding slope in the actual isothermal calorimetric curve, the accuracy of the transformation image of the modal transformation model is determined, thereby verifying the model effect and generating model verification results.

[0097] The overall hydration degree of the BSE test data within time t is compared with the corresponding overall hydration degree of the actual thermogravimetric results to determine the accuracy of the transformation image of the modal transformation model, thereby verifying the model effect and generating model verification results.

[0098] As an optional environment, the training environment for the modality transformation model is: Operating system: Ubuntu 20.04LTS; Model runtime environment: PyTorch + CUDA; Graphics card: NVIDIA 4090×4; Graphics card driver version: 520.56.06; CUDA version: 11.8; PyTorch version: Stable (1.13.1).

[0099] Step 214: Using the modal transformation model, perform modal transformation on the 3D imaging data to be transformed to generate BSE data.

[0100] Specifically, the 3D imaging data to be converted is input into the modal conversion model, which converts the CT image data into BSE image data and outputs the converted BSE data.

[0101] During the forward diffusion process, T noise levels are added to transform the image into a Gaussian noise distribution. T is chosen to be 2000. During training, a large T value does not pose a problem for the model, as the training process mainly involves the denoising model restoring the image from step t to step (t-1). However, during image prediction, T iterations are required to restore the low-resolution hydrolyzed image to a high-resolution one. Therefore, in this study, the inference process requires running the convolutional neural network model 2000 times, which significantly increases the time required to generate the hydrolyzed image.

[0102] The slow prediction speed of diffusion models is due to their dependence on Markov chains in the inference process. To address this issue, the model training primarily relies on marginal distributions. It does not directly depend on the joint distribution. During the inference process, Markov chain dependencies are broken by accelerating sampling, thereby improving prediction speed.

[0103] The specific prediction process is as follows:

[0104] in, In step 210, according to formula Reconstructed x0; The noise predicted by the convolutional neural network model at step t; It is the correlation coefficient of the Markov chain; β represents the proportion of data retained in step t-1. t-1 Let be the noise level at step t-1; This represents the proportion of data retained after continuous noise addition from the start to time t-1. These are the model's internal parameters; The BSE data is at time t-1; It is noise.

[0105] By reducing noise terms This can further accelerate the inference process and allow for direct prediction of states closer to the target distribution at each step, gradually reducing the number of iterations.

[0106] It is worth noting that the acquisition, storage, use, and processing of data in the technical solution of this application all comply with relevant laws and regulations. The user information in the embodiments of this application was obtained through legal and compliant means, and the acquisition, storage, use, and processing of user information have been authorized and agreed upon by the client.

[0107] It is worth noting that the information collected in this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0108] It is worth noting that the technical solution provided in this application provides users with a corresponding operation entry point, allowing users to choose to agree to or reject the automated decision-making results; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0109] This invention utilizes deep learning to convert cement hydration CT images into BSE images, breaking through the limitations of traditional imaging characterization techniques and achieving four-dimensional characterization under high resolution and high contrast conditions. It can provide imaging process support for basic theoretical research on cement hydration and provide non-destructive testing optimization methods for the performance evolution and durability issues of cement-based materials.

[0110] The technical solution of the modal conversion method for cement hydration image data provided in this invention involves preparing a hardened cement slurry sample; scanning the hardened cement slurry sample to obtain three-dimensional imaging sample data and electron microscope scanning sample data of cement hydration; constructing a sample dataset based on the three-dimensional imaging sample data and electron microscope scanning sample data, the sample dataset including a training dataset; training a preset diffusion model using the training dataset to construct a modal conversion model; and performing modal conversion on the three-dimensional imaging data to be converted using the modal conversion model to generate electron microscope scanning data. This achieves the modal conversion of X-CT images of cement hydration microstructures to BSE images, effectively combining the temporal-spatial imaging advantages of X-CT with the high resolution and contrast advantages of BSE, thereby significantly improving the spatial resolution of hydration images, ensuring evolution imaging across the entire time scale, comprehensively reflecting the entire process of cement hydration, and reducing costs. Simultaneously, it effectively distinguishes hydration products with similar densities, accurately captures details in the hydration process, and achieves four-dimensional imaging of the cement hydration process.

[0111] Figure 3 This is a schematic diagram of a modal conversion device for cement hydration image data provided in an embodiment of the present invention. This device is used to perform the aforementioned modal conversion method for cement hydration image data, such as... Figure 3 As shown, the device includes: a cement slurry sample preparation unit 11, a sample data acquisition unit 12, a sample dataset construction unit 13, a model training unit 14, and a modality conversion unit 15.

[0112] The cement slurry sample preparation unit 11 is used to prepare hardened cement slurry samples.

[0113] The sample data acquisition unit 12 is used to acquire three-dimensional imaging sample data and electron microscope scanning sample data of cement hydration by scanning hardened cement slurry samples.

[0114] The sample dataset construction unit 13 is used to construct a sample dataset based on three-dimensional imaging sample data and electron microscope scanning sample data. The sample dataset includes a training dataset.

[0115] The model training unit 14 is used to train a preset diffusion model using a training dataset to build a modality transformation model.

[0116] The modal conversion unit 15 is used to perform modal conversion on the three-dimensional imaging data to be converted through the modal conversion model, and generate electron microscope scanning data.

[0117] In this embodiment of the invention, the cement slurry sample preparation unit 11 is specifically used to prepare hardened cement slurry at different ages; the hardened cement slurry is fixed and sliced ​​to obtain hardened cement slurry samples.

[0118] In this embodiment of the invention, the sample data acquisition unit 12 is specifically used to scan and image the hardened cement slurry sample using computed tomography (CT) technology to generate three-dimensional imaging sample data; to perform carbon spraying on the hardened cement slurry sample after CT scanning, and to scan the carbon-sprayed hardened cement slurry sample using electron microscopy backscattering technology to generate electron microscopy scanning sample data.

[0119] In this embodiment of the invention, the sample dataset construction unit 13 is specifically used to establish a sample mapping relationship based on three-dimensional imaging sample data and electron microscope scanning sample data. The sample mapping relationship includes three-dimensional imaging sample data and corresponding electron microscope scanning sample data. Through an image registration algorithm, spatial registration and distortion correction are performed on the three-dimensional imaging sample data and the corresponding electron microscope scanning sample data to generate paired sample data. Based on multiple pairs of paired sample data, a sample dataset is constructed. The sample dataset is divided into a training dataset and a validation dataset.

[0120] In this embodiment of the invention, the device further includes an augmentation unit 16.

[0121] Augmentation unit 16 is used to perform data augmentation processing on the training dataset to generate an augmented training dataset. The data augmentation processing includes rotation, region cutting, random flipping, random grayscale adjustment, and random brightness adjustment.

[0122] In this embodiment of the invention, the modality conversion unit 15 is specifically used to use electron microscope scanned sample data as a supervision signal, perform forward diffusion on the electron microscope scanned sample data according to a preset noise signal at a preset time step to generate noise image data; perform reverse generation on the noise image data according to a preset time step to obtain a predicted noise signal; use the three-dimensional imaging sample data corresponding to the electron microscope scanned sample data as a constraint condition, perform condition generation according to the predicted noise signal and the noise image data to obtain the restored electron microscope scanned data; and perform backpropagation based on the loss result between the restored electron microscope scanned data and the electron microscope scanned sample data through a loss function to iteratively update the model parameters and construct a modality conversion model.

[0123] In this embodiment of the invention, the device further includes a model testing unit 17 and a model verification unit 18.

[0124] The model testing unit 17 is used to test the modality transformation model using the constructed test dataset to obtain electron microscope scanning test data.

[0125] The model validation unit 18 is used to compare the volume fraction based on electron microscope scanning test data and actual measurement data, and generate model validation results.

[0126] In this embodiment of the invention, a hardened cement slurry sample is prepared; the hardened cement slurry sample is scanned to obtain three-dimensional imaging sample data and electron microscope scanning sample data of cement hydration; a sample dataset is constructed based on the three-dimensional imaging sample data and electron microscope scanning sample data, the sample dataset including a training dataset; a preset diffusion model is trained using the training dataset to construct a modal conversion model; the modal conversion model is used to perform modal conversion on the three-dimensional imaging data to be converted, generating electron microscope scanning data, thus realizing the modal conversion of X-CT images of cement hydration microstructure to BSE images. This effectively combines the temporal-spatial imaging advantages of X-CT with the high resolution and contrast advantages of BSE, thereby significantly improving the spatial resolution of hydration images, ensuring evolution imaging across the entire time scale, comprehensively reflecting the entire process of cement hydration, and reducing costs; at the same time, it effectively distinguishes hydration products with similar densities, accurately captures details in the hydration process, and realizes four-dimensional imaging of the cement hydration process.

[0127] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer device, specifically, a computer device can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0128] This invention provides a computer device, including a memory and a processor. The memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions. When the program instructions are loaded and executed by the processor, they implement the steps of the above-described embodiment of the modal conversion method for cement hydration image data. For a detailed description, please refer to the above-described embodiment of the modal conversion method for cement hydration image data.

[0129] The following is for reference. Figure 4 It shows a schematic diagram of the structure of a computer device 600 suitable for implementing the embodiments of this application.

[0130] like Figure 4As shown, the computer device 600 includes a central processing unit (CPU) 601, which can perform various appropriate tasks and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage section 608 into random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the computer device 600. The CPU 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0131] The following components are connected to I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal feedback (LCD), etc., and speakers, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to I / O interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 610 as needed so that computer programs read from it can be installed in storage section 608 as needed.

[0132] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program including program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611.

[0133] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0134] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0135] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0136] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0138] It should also be noted that 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. Without further limitation, 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 said element.

[0139] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0140] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0141] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0142] This application can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0143] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0144] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

Claims

1. A method for modal conversion of cement hydration image data, characterized in that, The method includes: Prepare hardened cement slurry samples; By scanning the hardened cement slurry sample, three-dimensional imaging sample data and electron microscope scanning sample data of cement hydration were obtained; Based on the three-dimensional imaging sample data and electron microscope scanning sample data, a sample dataset is constructed, specifically including: Based on the three-dimensional imaging sample data and the electron microscope scanning sample data, a sample mapping relationship is established, which includes the three-dimensional imaging sample data and the corresponding electron microscope scanning sample data. Using an image registration algorithm, spatial registration and distortion correction are performed on the three-dimensional imaging sample data and the corresponding electron microscope scan sample data to generate paired sample data; Construct a sample dataset based on multiple pairs of paired sample data; The sample dataset is divided into a training dataset and a validation dataset; Using the training dataset, a pre-defined diffusion model is trained to construct a modality transformation model, specifically including: Using the electron microscope scanned sample data as a monitoring signal, the electron microscope scanned sample data is forward diffused according to a preset noise signal at a preset time step to generate noise image data. According to a preset time step, the predicted noise signal is generated in reverse based on the noise image data. Using the three-dimensional imaging sample data corresponding to the electron microscope scanning sample data as constraints, condition generation is performed based on the predicted noise signal and noise image data to obtain the restored electron microscope scanning data. The modality transformation model is constructed by backpropagating based on the loss result between the restored electron microscope scanning data and the electron microscope scanning sample data using a loss function, iterating and updating the model parameters. The modal transformation model is used to perform modal transformation on the three-dimensional imaging data to be transformed, generating electron microscope scanning data.

2. The modal conversion method for cement hydration image data according to claim 1, characterized in that, The preparation of the hardened cement slurry sample includes: Prepare hardened cement slurry at different ages; The hardened cement slurry was fixed and sliced ​​to obtain a hardened cement slurry sample.

3. The modal conversion method for cement hydration image data according to claim 1, characterized in that, The process of scanning the hardened cement slurry sample to obtain three-dimensional imaging sample data and electron microscopy scanning sample data of cement hydration includes: The hardened cement slurry sample is scanned and imaged using computed tomography (CT) technology to generate the three-dimensional imaging sample data. The hardened cement slurry sample after computed tomography (CT) scan was subjected to carbon spraying, and the hardened cement slurry sample after carbon spraying was scanned by electron microscopy backscattering technology to generate the electron microscopy scan sample data.

4. The modal conversion method for cement hydration image data according to claim 1, characterized in that, Before training the preset diffusion model using the training dataset to construct the modality transformation model, the method further includes: The training dataset is augmented to generate an augmented training dataset. The augmentation process includes rotation, region segmentation, random flipping, random grayscale adjustment, and random brightness adjustment.

5. The modal conversion method for cement hydration image data according to claim 1, characterized in that, The method further includes: The modality conversion model was tested using the constructed test dataset to obtain electron microscope scanning test data. The volume fraction is compared with the electron microscope scanning test data and the actual measurement data to generate model verification results.

6. A modal conversion device for cement hydration image data, characterized in that, The device includes: Cement slurry sample preparation unit, used to prepare hardened cement slurry samples; The sample data acquisition unit is used to acquire three-dimensional imaging sample data and electron microscope scanning sample data of cement hydration by scanning the hardened cement slurry sample; The sample dataset construction unit is used to construct a sample dataset based on the three-dimensional imaging sample data and the electron microscope scanning sample data; The model training unit is used to train a pre-defined diffusion model using a training dataset to build a modality transformation model. The modality conversion unit is used to perform modality conversion on the three-dimensional imaging data to be converted through the modality conversion model, and generate electron microscope scanning data; The sample dataset construction unit is specifically used to establish a sample mapping relationship based on the three-dimensional imaging sample data and the electron microscope scanning sample data. The sample mapping relationship includes the three-dimensional imaging sample data and the corresponding electron microscope scanning sample data. Through an image registration algorithm, spatial registration and distortion correction are performed on the three-dimensional imaging sample data and the corresponding electron microscope scanning sample data to generate paired sample data. Based on multiple pairs of paired sample data, a sample dataset is constructed. The sample dataset is divided into a training dataset and a validation dataset. The model training unit is specifically used to: use the electron microscope scanned sample data as a supervision signal; perform forward diffusion on the electron microscope scanned sample data according to a preset noise signal at a preset time step to generate noisy image data; perform reverse generation based on the noisy image data according to a preset time step to obtain a predicted noise signal; use the three-dimensional imaging sample data corresponding to the electron microscope scanned sample data as a constraint condition; perform conditional generation based on the predicted noise signal and the noisy image data to obtain the restored electron microscope scanned data; and perform backpropagation based on the loss result between the restored electron microscope scanned data and the electron microscope scanned sample data through a loss function to iteratively update the model parameters and construct the mode conversion model.

7. A computer-readable medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the modal conversion method for cement hydration image data as described in any one of claims 1 to 5.

8. A computer device comprising a memory and a processor, the memory for storing information including program instructions, and the processor for controlling the execution of the program instructions, characterized in that, When the program instructions are loaded and executed by the processor, the modal conversion method for cement hydration image data as described in any one of claims 1 to 5 is implemented.

9. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the modal conversion method for cement hydration image data as described in any one of claims 1 to 5.

Images