Method and system for identifying cement micro-constituents and hydration progress

Abstract

The present disclosure provides a cement micro-component and hydration process identification method and system, relating to the technical field of cement component identification, comprising obtaining a low-magnification SEM image and a high-magnification SEM image of a cement-based material; inputting the low-magnification SEM image and the high-magnification SEM image into a dual-channel micro-component segmentation network, extracting global features of the low-magnification SEM image through a low-channel network branch, and extracting local features of the high-magnification SEM image through a high-channel network branch; fusing the global features and the local features in a multi-head self-attention mechanism to obtain a fused feature map; performing semantic segmentation using the fused feature map to obtain a multi-scale component segmentation image of the cement-based material; extracting multi-feature data of an unhydrated particle area ratio, a hydration product area ratio, and an average pore diameter from the multi-scale component segmentation image, inputting the multi-feature data into a process classification model, and outputting a hydration process class label to realize classification and judgment of the hydration stage of the cement-based material.

Description

Technical Field

[0001] This disclosure relates to the field of cement component identification technology, specifically to a method and system for determining the microstructure and hydration process of cement. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] Cement-based materials are widely used in the fields of construction and civil engineering. The type of their microstructure and the hydration process directly affect the strength, durability, and other key properties of the materials. Therefore, accurate identification of the microstructure of cement and dynamic assessment of the hydration process are of great significance.

[0004] Traditional cement microstructure analysis and hydration process assessment rely on manual identification of cement microstructures in scanning electron microscope (SEM) images, such as unhydrated particles like C3S, C2S, C3A, and C4AF, and hydration products like CSH, CH, and ettringite. This method is not only inefficient but also requires a high level of expertise and experience from the observer, and it is difficult to ensure consistency and accuracy when identifying complex microstructures.

[0005] In recent years, the development of deep learning technology has provided strong support for image recognition and classification tasks. By building neural network models, it is possible to train on a large number of datasets, thereby accurately identifying the microscopic components of cement in SEM images.

[0006] However, the microstructure of cement contains both large-scale macroscopic information (such as particle shape and pore distribution) and microscopic details (such as the micromorphology of hydration products). Existing methods typically use single-magnification SEM images for analysis, which cannot comprehensively capture the multi-scale information of cement microstructure, resulting in insufficient accuracy in determining cement composition and hydration stage. Summary of the Invention

[0007] To address the aforementioned issues, this disclosure proposes a method and system for discriminating the microstructure and hydration process of cement. By designing a dual-channel convolutional neural network to receive and process low-magnification and high-magnification images respectively, a multi-head self-attention mechanism is introduced to improve the accuracy of feature maps. The method calculates key features such as the area ratio of unhydrated particles, the area ratio of hydration products, and the average pore diameter to classify and judge the hydration stage, thereby describing the hydration stage of cement materials and achieving efficient and intelligent judgment of the hydration process.

[0008] According to some embodiments, the present disclosure adopts the following technical solutions:

[0009] A method for determining the microstructure and hydration process of cement includes:

[0010] Acquire low-magnification and high-magnification SEM images of cement-based materials and preprocess them;

[0011] The preprocessed low-magnification SEM images and high-magnification SEM images are respectively input into a dual-channel micro-component segmentation network. The global features of the low-magnification SEM image are extracted through the low-channel network branch, and the local features of the high-magnification SEM image are extracted through the high-channel network branch.

[0012] Global and local features are fused using a multi-head self-attention mechanism to obtain a fused feature map;

[0013] Semantic segmentation was performed using fused feature maps to obtain multi-scale component segmentation images of cement-based materials;

[0014] Multi-feature data, including the area ratio of unhydrated particles, the area ratio of hydration products, and the average pore diameter, are extracted from multi-scale component segmentation images. This multi-feature data is then input into a process classification model, which outputs hydration process category labels to classify and determine the hydration stage of cement-based materials.

[0015] According to some embodiments, the present disclosure adopts the following technical solutions:

[0016] A cement microstructure and hydration process discrimination system includes:

[0017] The image acquisition module is used to acquire low-magnification and high-magnification SEM images of cement-based materials and perform preprocessing.

[0018] The feature extraction module is used to input the preprocessed low-magnification SEM image and high-magnification SEM image into the dual-channel micro-component segmentation network, respectively. The low-channel network branch extracts the global features of the low-magnification SEM image, and the high-channel network branch extracts the local features of the high-magnification SEM image.

[0019] The feature fusion module is used to fuse global and local features using a multi-head self-attention mechanism to obtain a fused feature map.

[0020] The component segmentation module is used to perform semantic segmentation using fused feature maps to obtain multi-scale component segmentation images of cement-based materials.

[0021] The process discrimination module is used to extract multi-feature data such as the area ratio of unhydrated particles, the area ratio of hydration products, and the average pore diameter from multi-scale component segmentation images. The multi-feature data is input into the process classification model and outputs hydration process category labels to realize the classification and judgment of the hydration stage of cement-based materials.

[0022] According to some embodiments, the present disclosure adopts the following technical solutions:

[0023] A computer program product includes a computer program that, when executed by a processor, implements the method for determining the microstructure and hydration process of cement.

[0024] According to some embodiments, the present disclosure adopts the following technical solutions:

[0025] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned method for determining the microstructure and hydration process of cement.

[0026] According to some embodiments, the present disclosure adopts the following technical solutions:

[0027] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the method for determining the microstructure and hydration process of cement.

[0028] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0029] This disclosure discloses a method for identifying the microstructure and hydration process of cement. It utilizes multi-scale information from low-magnification (100-1000x) and high-magnification (1000-10000x) SEM images, extracting features through a dual-channel convolutional neural network. The low-magnification image channel is designed to capture macroscopic features, employing deeper convolutional layers and large-pooling kernels to aggregate large-scale structural information such as particles and pores. The high-magnification image channel focuses on microscopic details, using shallower convolutional layers and small-pooling kernels to precisely capture the microscopic features of hydration products. By fusing information from both scales, the model can comprehensively determine the microstructure of cement, effectively improving the accuracy and stability of identification.

[0030] This disclosure discloses a method for discriminating the microstructure and hydration process of cement. It introduces a multi-head self-attention mechanism to correlate features extracted from low-magnification and high-magnification image channels. The self-attention mechanism generates Query, Key, and Value matrices for features from both low-magnification and high-magnification images, calculates feature weight matrices, and achieves adaptive feature enhancement, resulting in higher correlation between information from low-magnification and high-magnification channels during the fusion process. This not only strengthens the model's ability to distinguish cement microstructures but also improves the accuracy of identifying fine structures, ensuring the reliability of semantic segmentation results.

[0031] This disclosure discloses a method for discriminating the microstructure and hydration process of cement. It calculates key features such as the area ratio of unhydrated particles, the area ratio of hydration products, and the average pore diameter to describe the degree of hydration in cement materials. By inputting these features into a dynamic weighted random forest classification model, a time-series weight allocation algorithm is designed to dynamically adjust the decision path of the random forest. This allows the model to adjust the importance of features in real time according to the hydration process stages, thereby improving classification performance. The hydration process is divided into five stages: pre-induction, induction, acceleration, deceleration, and stabilization, achieving efficient and intelligent judgment of the hydration process. This method can effectively monitor the microscopic changes of cement at each stage, providing dynamic feedback support for subsequent material design and construction processes. Attached Figure Description

[0032] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0033] Figure 1 This is a schematic diagram of the residual block structure according to an embodiment of the present disclosure;

[0034] Figure 2 This is a flowchart of a method according to an embodiment of the present disclosure. Detailed Implementation

[0035] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0036] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0037] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0038] Example 1

[0039] One embodiment of this disclosure provides a method for determining the microstructure and hydration process of cement, including the following steps:

[0040] Step 1: Obtain low-magnification and high-magnification SEM images of cement-based materials and preprocess them;

[0041] Step 2: Input the preprocessed low-magnification SEM image and high-magnification SEM image into the dual-channel micro-component segmentation network respectively. Extract the global features of the low-magnification SEM image through the low-channel network branch and extract the local features of the high-magnification SEM image through the high-channel network branch.

[0042] Step 3: Fuse global and local features using a multi-head self-attention mechanism to obtain a fused feature map;

[0043] Step 4: Use the fused feature map to perform semantic segmentation to obtain a multi-scale component segmentation image of the cement-based material;

[0044] Step 5: Extract multi-feature data such as the area ratio of unhydrated particles, the area ratio of hydration products, and the average pore diameter from the multi-scale component segmentation image. Input the multi-feature data into the process classification model and output the hydration process category label to realize the classification and judgment of the hydration stage of cement-based materials.

[0045] As one embodiment, the specific implementation process of the method for determining the microstructure and hydration process of cement disclosed herein is as follows:

[0046] Step 1: Obtain low-magnification and high-magnification SEM images of cement-based materials and preprocess them;

[0047] Cement-based material samples that have undergone different hydration times were collected and scanned using SEM. Low-magnification and high-magnification SEM images of each cement sample were acquired one by one to construct a dataset. The low-magnification range was 100-1000x, and the high-magnification range was 1000-10000x.

[0048] Furthermore, the low-magnification and high-magnification SEM images of each cement sample were preprocessed, including data augmentation (rotation, flipping), size adjustment, brightness adjustment, etc.

[0049] Furthermore, the dataset is divided into a training set and a test set. Specifically, the training set accounts for 70% of the total dataset, and the test set accounts for 30%. Labels are added to the training set for neural network model training. The labels include annotations for unhydrated particles C3S, C2S, C3A, and C4AF; hydration products CSH and CH; ettringite; and pores.

[0050] Step 2: Construct and train a dual-channel micro-component segmentation network.

[0051] Specifically, the dual-channel micro-component segmentation network is a dual-channel convolutional neural network, comprising a low-channel network branch and a high-channel network branch, used to receive low-magnification and high-magnification images, respectively. The low-channel network branch (low-magnification image channel) is suitable for deeper convolutional layers and larger pooling kernels, used to extract large-scale global features (macro-level features), while the high-channel network branch (high-magnification image channel) is suitable for shallower convolutional layers and smaller pooling kernels, used to extract detailed local features.

[0052] Each channel network branch is designed with two residual blocks, namely residual block 1 and residual block 2. Residual block 1 consists of two convolutional layers with 128 kernels and 3×3 convolutional windows. Residual block 2 consists of two convolutional layers with 256 kernels and 3×3 convolutional windows.

[0053] Furthermore, the low-magnification image channels are configured as follows: containing an initial convolutional layer with 64 convolutional kernels and a 7×7 convolutional window; residual block 1 and residual block 2.

[0054] The high-magnification image channels are set as follows: an initial convolutional layer with 64 convolutional kernels and a 3×3 convolutional window; residual block 1 and residual block 2.

[0055] Furthermore, a multi-head self-attention mechanism is introduced after the outputs of the low-magnification and high-magnification image channels to fuse the feature maps of the low-magnification and high-magnification image channels. This includes: extracting global and local features, adjusting the feature maps of the global and local features to the same size through interpolation, aligning them in the spatial dimension, introducing the multi-head attention mechanism after the two channel network branches, converting the two aligned feature maps into a sequence form and inputting them into the multi-head attention mechanism, multiplying the self-attention weight matrices of the global and local feature maps with their corresponding value matrices to obtain the output features after self-attention, and finally concatenating the output features after self-attention along the channel dimension to obtain the fused feature map. Using a deconvolutional layer or an upsampling layer, the fused feature map is restored to the size of the input image and semantic segmentation is performed to generate a multi-scale component segmentation image.

[0056] Specifically, feature maps of global features and local features are obtained after passing through low-magnification and high-magnification image channels, respectively. The feature map of global features is F. low The size is (B, C, H, W); the feature map of the local features is F. high The size is (B, C, H, W). Where B is the batch size, i.e. the number of images processed in a batch, C is the number of channels, which is 3 for RGB images, and H and W are the dimensions of the feature map space.

[0057] When Flow and F high When the spatial dimensions are different, interpolation is used to adjust them to the same size to ensure F. low and F high Alignment in the spatial dimension. To apply the self-attention mechanism, the feature map F needs to be converted into a sequence form, flattened into a two-dimensional matrix, where each row represents a feature vector at a location. The flattening operation is as follows: F l =reshape(F low ,(B,H,W,C)),F h =reshape(F high (B,H,W,C)). After flattening, F l ,F h The size becomes (B, H × W, C).

[0058] As one embodiment, the fusion process after inputting into a multi-head self-attention mechanism includes the following steps:

[0059] First, generate Query, Key, and Value matrices from the feature maps of global features and local features, respectively:

[0060] Q l Q h =W q F l W q F h

[0061] K l ,K h =W k F l W k F h

[0062] V l V h =W v F l W v F h

[0063] Among them, W q W k W v It is a learnable weight matrix used to generate Query, Key, and Value. After transformation, the dimensions of Query, Key, and Value are all (B, H×W, d), where d is the dimension of the attention head, which should be reasonably selected according to the complexity of the task. Typically, 4, 6, or 8 can be selected, and the value can be increased or decreased based on the performance of the model.

[0064] Perform a dot product between the query and key for each of the low-magnification and high-magnification image channels to compute the self-attention weight matrix:

[0065] The attention weights for the low-magnification image channels are as follows:

[0066]

[0067] Attention weights for high-magnification image channels:

[0068]

[0069] The self-attention weight matrices of the low-magnification and high-magnification channels are multiplied by their corresponding value matrices to obtain the output features after self-attention:

[0070] Low-magnification image channel output:

[0071] F l out =A l ·V l

[0072] High-magnification image channel output:

[0073]

[0074] Finally, the outputs of the low-magnification and high-magnification channels are concatenated along the channel dimension to obtain the fused feature map:

[0075]

[0076] Finally, deconvolutional or upsampling layers are used to restore the fused feature map to the size of the input image, and semantic segmentation is performed to generate a semantic segmentation map. The softmax function is used as the activation function.

[0077] The training process uses forward propagation to calculate the loss, and then backpropagation is used for optimization.

[0078] Finally, a well-trained dual-channel micro-component segmentation network is obtained, which can realize multi-scale information extraction and component segmentation of SEM images.

[0079] Step 3: Build and train the process classification model

[0080] Specifically, a dynamically optimized random forest is used as the process classification model. Since the proportion of unhydrated particles gradually decreases and the proportion of hydration products increases, while the pore volume gradually decreases, multiple feature data—including the area ratio of unhydrated particles, the area ratio of hydration products, and the average pore diameter—are extracted from multi-scale component segmentation images. This data is then standardized or normalized to ensure that all features are on the same order of magnitude, enhancing model performance. The dataset is divided, with 70% used as the training set and 30% as the test set. The training set is then labeled with data representing the hydration stages for model training. The hydration stage is divided into five phases: pre-induction, induction, acceleration, deceleration, and stabilization.

[0081] Furthermore, the contribution of features to stage classification varies at different hydration stages: the proportion of unhydrated particle area is more important in the early stages, gradually decreasing in importance as hydration progresses; the proportion of hydration product area shows the opposite trend; and the average pore diameter is most important during the acceleration phase, reflecting changes in the hydration rate. Based on the above analysis, a time-series weight allocation algorithm is designed to dynamically adjust the decision path of the random forest, enabling the model to adjust feature importance in real time according to the hydration process stage, thereby improving classification performance.

[0082] The process of constructing a dynamic weighted random forest model is as follows:

[0083] Hydration stage time stamping: A pseudo-timestamp is assigned to each sample data point to represent its relative temporal position within the hydration process. A reasonable time range is assigned to each hydration stage: Early induction S1 Induction period S2 Acceleration period S3(4, 12], deceleration period S4(12, 24], and stabilization period S5(24, 48). For the i-th sample belonging to a certain stage Sk, according to the time range of that stage (t) k-1 , t k Generate pseudo timestamp: t i =t k-1 +(t k -t k-1 )·r, where r is a random number in the range [0, 1].

[0084] Time-dependent weighting function definition: A time-dependent weighting function is defined for each feature to account for changes in feature importance over time. For the proportion of unhydrated particle area, an exponential decay method is used. For the area ratio of hydration products, a linear increase is adopted: ω2(t)=k2·t+b2; for the average pore diameter, a Gaussian function is adopted: Among them, τ1, k2, b2, μ3, and σ3 are key parameters, which are reasonably set based on prior knowledge and experimental experience.

[0085] Random sampling: Assuming the training set has N samples, randomly sample N times to form a subset of the dataset. The resulting subset is also N in size, but samples may be repeated. Repeat the sampling process to generate multiple distinct subsets. Each subset is used to train a decision tree, i.e., a tree in a random forest. The number of trees in the random forest needs to be set within a reasonable range.

[0086] Decision tree generation: For a subset of data, during the training phase, the decision tree starts from the root node, which contains all the features in the subset, and randomly selects feature X. i The decision tree splits nodes at a split point s. The splitting operation is as follows: for each data sample at a node, the decision tree splits nodes based on the selected feature X. i The split point s divides the sample into two parts. If the feature values ​​of the sample satisfy X i If the feature value of the sample is ≤s, then proceed to the left child node; if the feature value of the sample satisfies X i If the value is greater than s, then proceed to the right child node. Iterate through all possible split points on the child node and split until the stopping requirement is met, until a split point is found that minimizes the weighted Gini coefficient after the split.

[0087] The number of features selected is generally the square root of the total number of features. The requirements for stopping node splitting include the depth of the tree and the number of nodes, which need to be set within a reasonable range.

[0088] Introducing a time-dependent weighted Gini coefficient: defined as the mixing degree of samples in node D, used to measure the purity of the node, the formula is:

[0089]

[0090] Where, ω i (t) is the time-related weight of the feature, p i It represents the proportion of samples in a node that belong to category i.

[0091] The process of generating a random forest is as follows: For each tree, nodes are split according to the sampled features until the stopping requirement is met, and all trees together form a random forest.

[0092] Ensemble prediction training process: During the training and testing phase, when a new sample X is input... i In a random forest, each decision tree independently classifies the sample; that is, when a sample reaches the last leaf node along the tree structure, it outputs a class label. Assuming the splitting rules of a tree can be represented by T as a set of decision paths, then the predicted class for an input sample x is... It can be represented as:

[0093]

[0094] Where T(x) represents the path from the root node to the leaf node according to the splitting rules of the decision tree T, and the final output category is the majority category label contained in the leaf node.

[0095] In a random forest, each decision tree T j (where j = 1, 2, ..., M) will independently affect sample X i Perform classification to obtain a category prediction. Because each tree uses different samples and feature subsets during training, their prediction results may differ. The final predicted category of a random forest... It is decided by majority vote, and the formula is:

[0096]

[0097] Where 'c' represents the category label, i.e., the hydration stage category; and 'M' is the total number of decision trees. It is an indicator function, when argmax is 1 if true, otherwise 0; c This means finding the category c that receives the most votes.

[0098] After training, the classification and judgment of the hydration stage of cement-based materials can be achieved.

[0099] As one example, such as Figure 2 As shown, the application process of the cement microstructure and hydration process discrimination method disclosed in this invention is as follows:

[0100] Step 1) Obtain low-magnification and high-magnification SEM scan images of cement at different hydration stages;

[0101] Step 2) Input them into the dual-channel micro-component segmentation network respectively, wherein global features of low-magnification SEM images are extracted through the low-magnification image channel, and local features of high-magnification SEM images are extracted through the high-magnification image channel;

[0102] Step 3) The global and local features are fused using a multi-head self-attention mechanism to obtain a fused feature map.

[0103] Step 4) After restoring the image size through upsampling of the fused feature maps, semantic segmentation is performed to obtain a multi-scale component segmentation image;

[0104] Step 5) Extract the area ratio of unhydrated particles, the area ratio of hydration products, and the average pore diameter from the multi-scale component segmentation image, and input them into the random forest model for classification to realize the identification of cement hydration stage.

[0105] Example 2

[0106] One embodiment of this disclosure provides a cement microstructure and hydration process discrimination system, including:

[0107] The image acquisition module is used to acquire low-magnification and high-magnification SEM images of cement-based materials and perform preprocessing.

[0108] The feature extraction module is used to input the preprocessed low-magnification SEM image and high-magnification SEM image into the dual-channel micro-component segmentation network, respectively. The low-channel network branch extracts the global features of the low-magnification SEM image, and the high-channel network branch extracts the local features of the high-magnification SEM image.

[0109] The feature fusion module is used to fuse global and local features using a multi-head self-attention mechanism to obtain a fused feature map.

[0110] The component segmentation module is used to perform semantic segmentation using fused feature maps to obtain multi-scale component segmentation images of cement-based materials.

[0111] The process discrimination module is used to extract multi-feature data such as the area ratio of unhydrated particles, the area ratio of hydration products, and the average pore diameter from multi-scale component segmentation images. The multi-feature data is input into the process classification model and outputs hydration process category labels to realize the classification and judgment of the hydration stage of cement-based materials.

[0112] Example 3

[0113] One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method for determining the microstructure and hydration process of cement.

[0114] Example 4

[0115] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they implement the method for determining the microstructure and hydration process of cement.

[0116] Example 5

[0117] One embodiment of this disclosure provides an electronic device, including a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the method for determining the microstructure and hydration process of cement.

[0118] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. 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, create a machine 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.

[0119] 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.

[0120] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A method for identifying cement micro-constituents and hydration progress, characterized by, The application relates to a cement-based material hydration stage classification method. The method comprises the following steps: acquiring and preprocessing low-magnification SEM images and high-magnification SEM images of a cement-based material; The preprocessed low-magnification SEM images and high-magnification SEM images are respectively input into a double-channel micro-component segmentation network, global features of the low-magnification SEM images are extracted through a low-channel network branch, and local features of the high-magnification SEM images are extracted through a high-channel network branch; The global features and the local features are fused in a multi-head self-attention mechanism to obtain a fused feature map; The fused feature map is used for semantic segmentation to obtain a multi-scale component segmentation image of the cement-based material; Multi-feature data such as an unhydrated particle area ratio, a hydration product area ratio and an average pore diameter are extracted from the multi-scale component segmentation image and input into a process classification model to output a hydration process category label, thereby realizing classification and judgment of the hydration stage of the cement-based material.

2. The method for determining the microstructure and hydration process of cement as described in claim 1, characterized in that, The low-magnification range of the low-magnification SEM images is 100-1000 times, the high-magnification range of the high-magnification SEM images is 1000-10000 times, and the preprocessing of the low-magnification SEM images and the high-magnification SEM images comprises data enhancement, size adjustment and brightness adjustment.

3. The method for determining the microstructure and hydration process of cement as described in claim 1, characterized in that, The double-channel micro-component segmentation network comprises a low-channel network branch and a high-channel network branch, the low-channel network branch is suitable for deep convolutional layers and large pooling kernels, and large-scale global features are extracted, the high-channel network branch is suitable for shallow convolutional layers and small pooling kernels, and detailed local features are extracted, and two residual blocks are designed for each channel network branch.

4. The method for determining the microstructure and hydration process of cement as described in claim 1, characterized in that, After the global features and the local features are extracted, the feature map of the global features and the feature map of the local features are adjusted to the same size through interpolation operation, are aligned in the spatial dimension, the multi-head attention mechanism is introduced after the two channel network branches, the two aligned feature maps are converted into sequence form and input into the multi-head attention mechanism, the self-attention weight matrix of the feature map of the global features and the feature map of the local features is respectively multiplied by the corresponding Value matrix to obtain the output features after self-attention, and finally the output features after self-attention are spliced along the channel dimension to obtain the fused feature map, the deconvolutional layer or the up-sampling layer is used to restore the fused feature map to the size of the input image and perform semantic segmentation to generate the multi-scale component segmentation image.

5. The method of claim 1, wherein the method is characterized by: A dynamic weight random forest model is constructed, the constructed dynamic weight random forest model is used as the process classification model, the unhydrated particle area ratio, the hydration product area ratio and the average pore diameter are extracted from the multi-scale component segmentation image and input into the dynamic weight random forest model, and the hydration stage is classified and judged, and the hydration process is divided into five stages of pre-induction, induction, acceleration, deceleration and stability.

6. The method for determining the microstructure and hydration process of cement as described in claim 5, characterized in that, When the sample data of the area proportion of unhydrated particles, the area proportion of hydration products and the average pore diameter are input, each decision tree of the random forest independently classifies the sample, and when the sample reaches the last node, i.e., the leaf node, along the tree structure, a class label is output. The final output class is the majority class label contained in the leaf node. In the dynamic weight random forest, each decision tree independently classifies the sample x to obtain a class prediction, and the final prediction class of the dynamic weight random forest is determined by majority voting.

7. A system for identifying cement micro-constituents and hydration progress, characterized by The method comprises the following steps: An image acquisition module is configured to acquire low-magnification SEM images and high-magnification SEM images of the cement-based material and perform preprocessing; A feature extraction module is configured to input the preprocessed low-magnification SEM images and high-magnification SEM images into a dual-channel microcomponent segmentation network, extract global features of the low-magnification SEM images through a low-channel network branch, and extract local features of the high-magnification SEM images through a high-channel network branch; A feature fusion module is configured to fuse the global features and the local features in a multi-head self-attention mechanism to obtain a fused feature map; A component segmentation module is configured to perform semantic segmentation using the fused feature map to obtain a multi-scale component segmentation image of the cement-based material; A process discrimination module is configured to extract multi-feature data of the area proportion of unhydrated particles, the area proportion of hydration products and the average pore diameter from the multi-scale component segmentation image, input the multi-feature data into a process classification model, and output a hydration process class label to classify and determine the hydration stage of the cement-based material.

8. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the cement microcomponent and hydration process discrimination method of any one of claims 1-6.

9. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium is configured to store computer instructions, and the computer instructions are executed by the processor to implement the cement microcomponent and hydration process discrimination method of any one of claims 1-6.

10. An electronic device, comprising: The method comprises the following steps: A processor, a memory and a computer program are included, wherein the processor is connected with the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the cement microcomponent and hydration process discrimination method of any one of claims 1-6.

Images