Cross-modal image registration method and system for cement hydration microstructure analysis

Abstract

The present invention belongs to the field of image processing technology, and in particular relates to a cross-modal image registration method and system for cement hydration microstructure analysis; the method comprises: obtaining cement hydration microstructure images, including CT images and BSE images, and extracting significant phases from the images; using the processed BSE images as fixed images, performing affine transformation on the processed CT images, and iteratively optimizing the transformation parameters using a progressive competitive pyramid particle swarm optimization algorithm; the present invention aligns BSE images and CT images by extracting significant cement particle phase information, and can effectively solve the significant challenges of cement image homogeneity and multimodality; the provided registration framework can effectively solve the cross-modal registration problem of cement microstructure images, thereby realizing multi-source heterogeneous image analysis of cement hydration microstructure.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a cross-modal image registration method and system for cement hydration microstructure analysis. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Cement is an indispensable cementitious material in all kinds of construction. Its important position in human society stems from its excellent macroscopic properties, which are determined by its microstructure. Through multi-angle research, researchers have deeply explored the relationship between cement's microstructure, reaction mechanism and performance, aiming to design high-performance cement materials that meet social needs.

[0004] Cement microstructure images intuitively characterize the microstructure of cement. Researchers usually use micro-CT scanners and scanning electron microscopes (SEM) to observe its subtle features. Micro-CT penetrates the sample through X-rays. Due to the different absorption of X-rays by different hydration products, different grayscale distributions are produced in the image. These grayscale differences reflect the distribution of hydration products and thus present the internal microstructure of cement. Compared with micro-CT, the backscatter (BSE) image obtained in the backscatter mode of the scanning electron microscope (SEM) has a higher resolution, which can more finely observe the distribution of hydration products on the surface of cement samples.

[0005] However, due to the limitation of imaging mechanism, SEM can only obtain the distribution information of sample surface, but not the internal structure. Therefore, combining the images obtained by micro-CT and SEM can provide researchers with more complete internal structure and component distribution information, and further deepen the understanding of cement materials. In order to realize the information integration of the two images, the two types of images need to be aligned, which is a process of image registration.

[0006] To align the two types of cement microstructure images, it is necessary to deal with challenges such as the homogeneity of the cement microstructure itself and the inconsistency between CT images and BSE images in terms of modality, dimension, and resolution.

[0007] Among the above problems, the biggest challenge comes from the homogeneity of microstructure images. Cement microstructure images contain numerous blocky unhydrated cement particles of various sizes and shapes and hydration products of amorphous shapes. Such a structure makes the result after the transformation image still have a large similarity with the previous image, which is manifested in the presence of many local optimal solutions in the optimized topographic map, making it difficult for the optimization algorithm to find better registration parameters.

[0008] Conventional feature point-based registration algorithms, such as SIFT (Scale Invariant Feature Transform), SURF (Speeded Up Robust Features), FAST (Speeded Up Segmentation Test Features) and other classic matching algorithms usually rely on image intensity changes to extract feature point descriptions. However, the two types of image modalities are inconsistent and the intensity changes are quite different, resulting in feature changes in one image that cannot be applied to the other modality. In addition, homogeneous structural textures make the above algorithms find a large number of feature points, some of which may be similar in spatial distance, but the final matching results are far apart.

[0009] Therefore, extracting and utilizing significant information (such as significant particles) to overcome the homogeneity problem of cement hydration microstructure images and then realizing cement hydration microstructure image registration based on significant information is a solution with significant potential. However, due to the wide search range of registration parameters, the solution space is very large. Conventional swarm intelligence optimization methods have low search efficiency, are prone to premature maturity, and are prone to falling into local optimality, which is not conducive to finding registration parameters. Summary of the invention

[0010] In order to solve the deficiencies of the prior art, the present invention provides a cross-modal image registration method, system, electronic device, computer-readable storage medium and computer program product for cement hydration microstructure analysis; first, in order to cope with the homogeneity challenge and multimodality challenge of cement microstructure images, the significant phases of the microstructure images are extracted by image contrast enhancement, image segmentation, morphological processing, statistical size of cement particles and filtering, and such images containing only significant phases can effectively solve the homogeneity and multimodality problems; then, by constructing a 2D / 3D cross-dimensional transformation model, the two types of images are mapped to the same dimension to overcome the problem of inconsistent dimensions of BSE images and CT images; finally, considering that the registration optimization parameter solution space has a wide range and the objective function is a multi-peak function, there are many local optimal solutions, and the conventional gradient-based continuous optimization algorithm and some intelligent optimization algorithms are prone to fall into the local optimal solution, a progressive competitive pyramid particle swarm optimization algorithm is proposed to optimize the registration transformation parameters; the improved algorithm not only has a powerful search capability to avoid falling into the local optimum in the process of searching for registration optimization parameters, but also has a faster convergence speed.

[0011] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0012] In a first aspect, the present invention provides a cross-modal image registration method for cement hydration microstructure analysis, comprising:

[0013] Acquire cement hydration microstructure images, including CT images and BSE images;

[0014] Extracting significant phases from the acquired image, including: using a contrast enhancement algorithm to enhance the image contrast, using an image segmentation method to segment cement particles and hydration products in the enhanced image; extracting significant cement particle microstructures from the segmentation results, and processing them using morphological operations; obtaining an image containing only significant cement particle phases by retaining cement particles whose areas reach a set threshold in the image;

[0015] Affine transformation is performed on the CT image and BSE image after significant phase extraction, including: taking the BSE image after significant phase extraction as a fixed image, taking the coordinate space of the CT image as a standard transformation space, and constructing a two-dimensional plane area with the same size as the BSE image based on the transformation parameters; transforming the two-dimensional plane area to obtain slices along a certain direction in the CT image, and taking the CT image slices as floating images matched with the BSE image; using the progressive competitive pyramid particle swarm optimization algorithm to iteratively optimize the transformation parameters to obtain the optimal transformation parameters to determine the CT image slice with the highest fitness with the BSE image;

[0016] The method uses a progressive competition pyramid particle swarm optimization algorithm to iteratively optimize the transformation parameters, including: dividing the optimization process of the transformation parameters into two stages, the first stage adopts a particle competition strategy, pairs particles in pairs, and divides them into winners and losers according to fitness, the losers learn from the winners, and the winners directly enter the next generation; the second stage divides the particles into stratified populations according to fitness, and the population adopts a pyramid hierarchical topological structure; the particles in each layer of the population compete with each other, the winners learn from the individual best in history and the particles in the previous layer of the population, and the losers learn from the winners of the same layer of the population.

[0017] In a second aspect, the present invention provides a cross-modal image registration system for cement hydration microstructure analysis, comprising:

[0018] The image acquisition and preprocessing module is configured to: acquire cement hydration microstructure images, including CT images and BSE images;

[0019] The image significant phase extraction module is configured to: extract significant phases from the acquired image, including: using a contrast enhancement algorithm to enhance the image contrast, using an image segmentation method to segment cement particles and hydration products in the enhanced image; extracting significant cement particle microstructures of the image from the segmentation results, and processing them using morphological operations; obtaining an image containing only significant cement particle phases by retaining cement particles whose areas reach a set threshold in the image;

[0020] The image registration module is configured to: perform affine transformation on the CT image and the BSE image after significant phase extraction, including: taking the BSE image after significant phase extraction as a fixed image, taking the coordinate space of the CT image as a standard transformation space, and constructing a two-dimensional plane area with the same size as the BSE image based on the transformation parameters; transforming the two-dimensional plane area to obtain slices along a certain direction in the CT image, and taking the CT image slices as floating images matched with the BSE image; iteratively optimizing the transformation parameters using a progressive competitive pyramid particle swarm optimization algorithm to obtain the optimal transformation parameters to determine the CT image slice with the highest fitness with the BSE image;

[0021] The method uses a progressive competition pyramid particle swarm optimization algorithm to iteratively optimize the transformation parameters, including: dividing the optimization process of the transformation parameters into two stages, the first stage adopts a particle competition strategy, pairs particles in pairs, and divides them into winners and losers according to fitness, the losers learn from the winners, and the winners directly enter the next generation; the second stage divides the particles into stratified populations according to fitness, and the population adopts a pyramid hierarchical topological structure; the particles in each layer of the population compete with each other, the winners learn from the individual best in history and the particles in the previous layer of the population, and the losers learn from the winners of the same layer of the population.

[0022] In a third aspect, the present invention provides an electronic device comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-mentioned cross-modal image registration method for cement hydration microstructure analysis.

[0023] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the steps of the above-mentioned cross-modal image registration method for cement hydration microstructure analysis.

[0024] In a fifth aspect, the present invention provides a computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned cross-modal image registration method for cement hydration microstructure analysis.

[0025] One or more of the above technical solutions have the following beneficial effects:

[0026] (1) The present invention takes into account the homogeneity of cement images and the modal differences of CT images and BSE images. Through contrast enhancement, image segmentation, and statistical cement particle size and filtering processing, two types of images that only retain significant cement particle phases are obtained. By using microstructure images that only contain significant phases for registration, the significant challenges brought by the homogeneity of cement images can be effectively overcome.

[0027] (2) The present invention provides an enhanced swarm intelligence optimization algorithm - progressive competitive pyramid particle swarm algorithm to optimize the transformation parameters in the registration process of cement microstructure images. The improved algorithm has powerful exploration capabilities and can effectively search the solution space to avoid falling into the local optimum in the process of searching for registration parameters.

[0028] (3) The present invention provides a cross-modal registration technology framework for cement microstructure images to achieve the linkage analysis of cement microstructure micro-CT images and BSE images. By extracting the significant phase information of the image and using images containing only significant phases for registration, the homogeneity of cement microstructure images is overcome. When constructing the transformation model, a CT image slice with the same size as the BSE image is determined based on the transformation parameters, and the CT image slice is used as a floating image, and the processed BSE image is used as a fixed image to achieve similarity comparison. Finally, the transformation parameters are iteratively optimized using a swarm intelligence optimization algorithm to ultimately obtain an accurate registration result. The registration framework of the present invention can effectively solve the cross-modal registration problem of cement microstructure images, realize multi-source heterogeneous image analysis of cement hydration microstructure, and thus promote the in-depth understanding and research of hydration microstructure and hydration mechanism.

[0029] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0031] Figure 1 This is a flow chart of a cross-modal image registration method for cement hydration microstructure analysis according to the first embodiment.

[0032] Figure 2 This is the first embodiment of the image salient phase extraction process.

[0033] Figure 3 The morphology of cement particles in the CT image of the first embodiment.

[0034] Figure 4 The morphology of cement particles in the BSE image of the first embodiment.

[0035] Figure 5 This is a flowchart of extracting significant phases from BSE images in the first embodiment.

[0036] Figure 6 This is an iterative optimization registration process of the first embodiment.

[0037] Figure 7 , Figure 8 , Fig. 9 They are the CT image, BSE image and registration result of sample 1 of the first embodiment respectively.

[0038] Fig.10 , Fig.11 , Fig.12 They are the CT image, BSE image and registration result of sample 2 of the first embodiment respectively.

[0039] Fig.13 , Fig.14 , Fig.15 They are the CT image, BSE image and registration result of sample 3 of the first embodiment respectively.

[0040] Fig.16 , Fig.17 , Fig.18 They are the CT image, BSE image and registration result of sample 4 of the first embodiment respectively. DETAILED DESCRIPTION

[0041] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0042] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0043] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0044] Embodiment 1

[0045] like Figure 1 As shown, this embodiment discloses a cross-modal image registration method for cement hydration microstructure analysis, comprising the following steps:

[0046] Step 1: Acquire cement hydration microstructure images, including CT images and BSE images;

[0047] Step 2: Extract the significant phases from the two types of images, unify the modalities of the two types of images while overcoming the homogeneity of cement hydration microstructure images;

[0048] Step 3: The BSE image after significant phase extraction is used as a fixed image, and the coordinate space of the CT three-dimensional image is used as the coordinate transformation space. CT image slices are constructed and affine transformation is performed on the slices. The transformation parameters are iteratively optimized using a swarm intelligence optimization algorithm to obtain the optimal transformation parameters to determine the CT image slice with the highest similarity to the BSE image.

[0049] Step 4: Apply the optimal transformation parameters obtained by iterative optimization to the original CT image for slicing, and finally achieve accurate matching of the two types of images.

[0050] In step 1, backscatter imaging is performed using a scanning electron microscope (SEM) to obtain a BSE image, and a three-dimensional CT image is obtained by micro-CT scanning.

[0051] In step 2, a preliminarily processed image is obtained by performing salient phase extraction on the original image to address the challenges of cross-modality and homogeneity.

[0052] like Figure 2 As shown in the figure, the process of extracting significant phases from images includes image contrast enhancement, image segmentation and area threshold control processing in sequence; first, the two types of images are enhanced using the CLAHE algorithm, and then the enhanced two types of images are segmented using the K-MEANS image segmentation algorithm, and the image containing cement particles is extracted through image segmentation, and the segmented image is morphologically processed, and finally the size of the connected area of ​​the image is calculated, and the significant cement particles are retained to obtain two types of images showing only significant phases; wherein, by counting the area of ​​cement particles in the image, the cement particles of different sizes are screened according to their area sizes, the smaller cement particles are eliminated, and only the larger significant cement particles are extracted to obtain the cement microstructure image containing the significant phase.

[0053] Step 2 specifically includes: Step 201: using the CLAHE contrast enhancement algorithm to perform image enhancement processing on the two types of images obtained respectively.

[0054] The contrast of the original BSE image and micro-CT image is low, which will affect the accuracy of image segmentation by the K-MEANS segmentation algorithm; therefore, the present invention uses the CLAHE algorithm to enhance the contrast of the image, improve the local image contrast and reduce the noise and over-enhancement problems caused by global contrast enhancement.

[0055] Step 202: Use the K-MEANS clustering algorithm to cluster the grayscales of the two types of images, thereby segmenting the cement particles and hydration products in the images.

[0056] In view of the multimodal and homogeneous characteristics of cement microstructure images, the present invention uses the inherent characteristics of cement microstructure images for processing. Specifically, in the microstructure images of cement hydration stages, there is a significant feature, namely, irregular blocky cement particles, which are distributed in different areas of the image and appear as areas with higher grayscale values ​​in both types of images, with high consistency; the morphology of cement particles in CT images and BSE images are as follows: Figure 3 , Figure 4 As shown, the red frame in the figure indicates block cement particles.

[0057] The present invention adopts K-MEANS image segmentation technology to extract these cement particles and retains larger cement particle blocks therein, thereby obtaining a processing result with significant physical phases.

[0058] CT images and BSE images have different modalities, and different segmentation parameters need to be set when using the K-MEANS algorithm for segmentation. The number of clusters in the K-MEANS segmentation method is determined by the image quality. If the physical phases in the microstructure image can be easily identified, more categories can be set. In this scheme, the BSE image is the cropped domain of interest and does not contain imaging other than the cement physical phase, while the CT image also contains a part of the unimaged area. Therefore, when dividing the segmentation categories, the number of BSE image categories is less than that of CT images. Although there are more than these categories of physical phases in cement, considering the noise of the image itself and the imaging quality problems, too detailed classification will bring difficulties to the subsequent matching, so it is not advisable to divide too many categories.

[0059] In both types of images, cement particles in the cement microstructure are areas with higher gray values, so an automatic segmentation algorithm based on K-MEANS is designed to extract the phase structure information of cement particles. It is only necessary to obtain the maximum clustering result of the algorithm clustering center during the clustering process, retain only the phase of this category, and exclude other phases to complete the extraction of cement particle phases.

[0060] Step 203: extracting the significant cement particle microstructure of the image from the segmentation result, and processing the image using morphological operations to obtain an image containing only significant cement particle phases;

[0061] Since the two types of images have noise in them, there are many scattered white noise points in the segmentation results, and there are many burrs on the edges of the objects. These problems need to be processed through some morphological processing, which can be alleviated through erosion and expansion operations. The erosion and expansion parameters of the two types of images are the same.

[0062] Step 204: extracting the contours of cement particles in the image, calculating the areas of cement particles according to the contours, and retaining cement particles in the image whose areas reach a set threshold.

[0063] After step 203, a cement microstructure image containing a significant phase is obtained, and then the size of the connected region in the cement particle phase is calculated, and cement particles that do not meet the threshold are filtered out to obtain an image containing only the significant phase.

[0064] The specific process is as follows: First, the outer contour of the significant object in the image and the holes inside it are found through the boundary tracking algorithm, and the contour is defined as a collection of a series of contour points. After obtaining the contour points of the object, the area of ​​each contour is calculated separately. The area calculation uses the polygon area formula, which is an efficient geometric algorithm used to calculate the area of ​​a polygon defined by a set of vertices. The algorithm principle is:

[0065] Assume that the contour consists of a series of vertices Definition, these points form a polygon, and the area is obtained by calculating the boundary points of the polygon; for a polygon consisting of The area of ​​a polygon composed of vertices can be calculated by the following formula:

[0066]

[0067] The area of ​​each cement particle in the image is obtained through the above calculation. The smaller cement particles are screened out and only the larger physical structure is retained, so that a microstructure image containing only significant cement particles can be obtained.

[0068] like Figure 5 The figure shows the flow chart of significant phase extraction of BSE images, including: obtaining the original BSE image, using CLAHE contrast to enhance the BSE image, using K-MEANS to segment the BSE image and perform morphological processing (erosion and expansion), and finally obtaining a microstructure image with significant cement particle phase by screening out smaller cement particles.

[0069] The present invention effectively realizes the physical phase extraction of the two types of images through the above method, alleviates the challenge brought by homogeneity to a certain extent, and also solves the problem of inconsistency of the two types of image modalities.

[0070] In step 3, a registration transformation model is established, and the transformation parameters of the model are optimized by using a progressive competitive pyramid particle swarm intelligent optimization algorithm. The optimal transformation parameters are obtained through iterative optimization.

[0071] like Figure 6As shown, the iterative optimization registration process includes: initializing the optimization parameters Tx, Ty, Tz, Rx, Ry, and Rz of the affine transformation model; slicing the CT image according to the current transformation parameters, calculating the similarity between the CT image slice and the BSE image, and judging whether the similarity is greater than a given threshold or reaches the maximum number of iterations; if so, slicing the original CT image according to the final parameters to obtain the matching result, outputting the matching image pair and the transformation parameters; if not, using the progressive competitive pyramid particle swarm optimization algorithm to iteratively optimize the transformation parameters. The specific steps include:

[0072] Step 301: Construct a 2D / 3D affine transformation model.

[0073] The affine transformation model is a cross-dimensional transformation model from a two-dimensional image to a three-dimensional image. The present invention actually compares a slice in a three-dimensional CT image. This is a 2D / 3D registration process, so it is necessary to define a transformation space and a transformation scheme; specifically, the following are included:

[0074] The BSE image after significant phase extraction is used as a fixed image; the transformation space is defined, and the coordinate space of the CT image is used as the standard transformation space;

[0075] A two-dimensional plane region with the same size as the BSE image is constructed based on the transformation parameters, corresponding to a slice of the CT image; each coordinate in the two-dimensional plane region corresponds to a voxel in the coordinate space corresponding to the CT image;

[0076] The transformation of the two-dimensional plane area corresponds to the slice transformation of the CT three-dimensional image. The voxels of the CT image can be directly indexed by the slice coordinates to obtain the slices along a certain direction in the CT image. The CT image slices are used as floating images that match the fixed image, thereby realizing the similarity comparison of cross-dimensional images.

[0077] Among them, affine transformation includes translation transformation and rotation transformation. The transformation order is to perform translation transformation first, and then rotate transformation around three coordinate axes (X, Y, Z coordinate axes). The transformation parameters include translation parameters and rotation parameters. The translation parameters are the displacement of the CT image slice relative to the origin of the CT image. The rotation parameters correspond to the rotation angle around the three coordinate axes, and the rotation center is the coordinate corresponding to the center position of the two-dimensional plane area.

[0078] Since the spatial resolution of the pixels of the two types of images is known during the extraction of significant objects in step 2, the high-resolution image is downsampled to be consistent with the low-resolution image to achieve a unified resolution. Unifying the spatial resolution of pixels can reduce the optimization of scaling parameters, so the affine transformation in this embodiment only includes translation and rotation transformations; if the spatial resolution of pixels of the two types of images is inconsistent, it is necessary to add scaling parameters of the affine transformation; the transformation order is to first perform scaling transformation, then translation transformation, and then rotation transformation around the three coordinate axes.

[0079] Specifically: The present invention creates an intermediate result, that is, the slice image obtained from the CT image is used as a floating image , BSE image as a fixed image , corresponds to a two-dimensional plane area in the coordinate space, and each point in the two-dimensional plane area corresponds to a pixel point in the CT image coordinate space;

[0080] Transforming a two-dimensional plane region is equivalent to floating it in the coordinate space corresponding to the CT image. The transformation result is a finite slice along a certain direction in the CT image, which is the same size as the BSE image.

[0081] The transformation of the floating image in the fixed image can be described as a rigid body transformation as follows:

[0082]

[0083] In the formula, Represents the coordinates of the floating image after transformation, Represents the coordinates of the floating image before transformation, represents the displacement transformation matrix, Represents the moving distance relative to the X, Y, and Z axes respectively. It is the rotation around the X, Y, and Z axes. Any rotation of a slice can be decomposed into matrix multiplication of rotations around the three axes.

[0084]

[0085]

[0086]

[0087] In the formula, is the rotation angle around the Z axis, is the rotation angle around the Y axis, is the rotation angle around the X axis.

[0088] Step 302: Use the intelligent optimization algorithm PC-PPSO to optimize the transformation parameters.

[0089] The traditional PSO algorithm has great limitations. When dealing with complex problems, there are problems of premature convergence and falling into local optimum. Therefore, the present invention improves the traditional PSO algorithm. The present invention uses a progressive competitive pyramid particle swarm optimization algorithm (Progressive Competitive - Pyramid Particle Swarm Optimization, PC-PPSO) to optimize the transformation parameters in the registration transformation process. The improved algorithm has a powerful search capability, which can avoid falling into the local optimum during the optimization process.

[0090] Step 3021: The core of the PC-PPSO algorithm lies in the position update and velocity update rules of the particles; the optimization of PC-PPSO is divided into two stages. The first stage focuses on exploration capability. Based on the entire population, all particles are directly paired with each other. Pairs of particles compete with each other. Particle pairs are divided into winners and losers according to their fitness. Losers learn from winners, and winners directly enter the next generation. In this way, the particle exploration performance can be enhanced, thereby fully searching the solution space.

[0091] After reaching a certain number of iterations, the algorithm enters the second stage. In this stage, the fitness of particles is sorted to establish a pyramid structure of a multi-layer population. Particles on the same layer compete with each other to produce winners and losers. Losers learn from winners and their best individual history, while winners learn from particles in the upper population and their best individual history. The winning particles in the top population are not updated and directly enter the next generation. The switching of stages is set by the ratio of the current number of iterations to the total number of iterations. The above update logic enables the algorithm to converge quickly while ensuring particle diversity, thereby improving the overall algorithm's exploration performance and accelerating the algorithm's convergence speed.

[0092] (1) The algorithm assumes that there are particles, the position of each particle represents the solution to the problem, which is a dimensional vector. The position of the global best particle is marked as , the best historical position of the i-th particle is marked as During the iteration process, the particles move with the speed , thereby optimizing.

[0093]

[0094]

[0095]

[0096] Here, i represents the i-th particle in a certain number of iterations. In the first stage, the winning particle is not updated and directly enters the next generation, while the failed particles learn from the winning particles. The update rules for the speed and position of the failed particles are:

[0097]

[0098]

[0099] In the formula, represents the number of iterations, Representative Substitute the moving speed of the i-th failed particle. , , , They represent the positions and speeds of the winner and loser in the k-th round of competition in the t-th generation, respectively. and is a random number between [0,1], which is used to increase the randomness of particle movement and thus increase the diversity of particles in the population.

[0100] (2) In the second stage of the PC-PPSO algorithm, the main focus is on ensuring particle diversity while accelerating the convergence of particles. The algorithm uses a pyramid hierarchical topology structure to update the particle speed.

[0101] First, the particles are sorted according to their fitness and multiple levels are set. The fitness of particles in the upper layer is greater than that of particles in the lower layer. The number of particles in each layer is not fixed, and usually increases from top to bottom.

[0102] Located in layer The particle velocity is updated as follows:

[0103]

[0104]

[0105]

[0106]

[0107] in, is a random number uniformly distributed within [0,1]. represent the j-th particle, the loser and the winner of the i-th layer respectively. They respectively represent the position, velocity and historical optimal position of the t-th generation particle p. Represents a random positive integer not greater than the number of particles in a certain layer.

[0108] (3) A hyperparameter is used to control the switching between different stages, and its switching state is related to the number of evaluations. It is defined as follows:

[0109]

[0110] In the formula, is the current evaluation number, is the number of evaluations to be terminated. When r reaches the set parameter, the particle update logic can be switched.

[0111] Step 3022: The purpose of image registration is to find a set of transformation parameters that can make the reference image and the floating image as similar as possible; this is equivalent to finding the transformation parameters that satisfy the maximum similarity function S in the optimization process of the PC-PPSO algorithm to determine the CT image slice with the highest similarity to the BSE image;

[0112] The similarity function is:

[0113]

[0114]

[0115] Where F represents the fixed image, M represents the floating image, T represents the spatial transformation, and S represents the similarity between the fixed image and the floating image.

[0116] The similarity function of the present invention uses the DICE coefficient, which is a common indicator for measuring the degree of overlap between the segmentation result and the ground truth. The two types of images have been segmented in the process of image salient object extraction, so this indicator can be used for comparison.

[0117] The definition of the DICE coefficient is as follows:

[0118]

[0119] In the formula, represents the number of elements of image A, is the number of elements of image B. Represents the intersection area of ​​image A and image B, which represents the overlapping part of the two. The DICE coefficient measures the degree of overlap between the predicted area and the true area by calculating the ratio of the intersection area of ​​the predicted area and the true area to their total area.

[0120] The above optimization process is to find the optimal transformation through iteration The transformation parameters corresponding to the floating image that is similar to the fixed image are found as much as possible; the transformation parameters correspond to a slice of the CT image, and the slice is compared with the processed BSE image for similarity. The fitness function of the comparison is the DICE coefficient. The transformation parameters are continuously iterated to optimize the DICE coefficient, thereby achieving the maximum overlap and similarity between the two images.

[0121] In step 4, it includes: applying the optimal transformation parameters after iterative transformation to slice the original CT image, obtaining the CT image slices corresponding to the BSE image, and outputting the matching slice results.

[0122] This example was experimented on 4 cement samples; Figure 7 , Figure 8 , Fig. 9 They are the CT image, BSE image and registration result of sample 1; Fig.10 , Fig.11 , Fig.12 They are the CT image, BSE image and registration result of sample 2; Fig.13 , Fig.14 , Fig.15 They are the CT image, BSE image and registration result of sample 3; Fig.16 , Fig.17 , Fig.18 They are the CT image, BSE image and registration result of sample 4 respectively.

[0123] Embodiment 2

[0124] This embodiment discloses a cross-modal image registration system for cement hydration microstructure analysis, including:

[0125] The image acquisition and preprocessing module is configured to: acquire cement hydration microstructure images, including CT images and BSE images;

[0126] The image significant phase extraction module is configured to: extract significant phases from the acquired image, including: using a contrast enhancement algorithm to enhance the image contrast, using an image segmentation method to segment cement particles and hydration products in the enhanced image; extracting significant cement particle microstructures of the image from the segmentation results, and processing them using morphological operations; obtaining an image containing only significant cement particle phases by retaining cement particles whose areas reach a set threshold in the image;

[0127] The image registration module is configured to: perform affine transformation on the CT image and the BSE image after significant phase extraction, including: taking the BSE image after significant phase extraction as a fixed image, taking the coordinate space of the CT image as a standard transformation space, and constructing a two-dimensional plane area with the same size as the BSE image based on the transformation parameters; transforming the two-dimensional plane area to obtain slices along a certain direction in the CT image, and taking the CT image slices as floating images matched with the BSE image; iteratively optimizing the transformation parameters using a progressive competitive pyramid particle swarm optimization algorithm to obtain the optimal transformation parameters to determine the CT image slice with the highest fitness with the BSE image;

[0128] The method uses a progressive competition pyramid particle swarm optimization algorithm to iteratively optimize the transformation parameters, including: dividing the optimization process of the transformation parameters into two stages, the first stage adopts a particle competition strategy, pairs particles in pairs, and divides them into winners and losers according to fitness, the losers learn from the winners, and the winners directly enter the next generation; the second stage divides the particles into stratified populations according to fitness, and the population adopts a pyramid hierarchical topological structure; the particles in each layer of the population compete with each other, the winners learn from the individual best in history and the particles in the previous layer of the population, and the losers learn from the winners of the same layer of the population.

[0129] It should be noted that the above modules correspond to the steps in Embodiment 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Embodiment 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer executable instructions.

[0130] Embodiment 3

[0131] Embodiment 3 of the present invention provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the above-mentioned cross-modal image registration method for cement hydration microstructure analysis are completed.

[0132] Embodiment 4

[0133] Embodiment 4 of the present invention provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps of the above-mentioned cross-modal image registration method for cement hydration microstructure analysis are completed.

[0134] Embodiment 5

[0135] Embodiment 5 of the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned cross-modal image registration method for cement hydration microstructure analysis.

[0136] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0137] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0138] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0139] The description of each embodiment in the above embodiments has different emphases. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0140] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A cross-modal image registration method for cement hydration microstructure analysis, characterized in that: include: Acquire cement hydration microstructure images, including CT images and BSE images; Extracting significant phases from the acquired image, including: using a contrast enhancement algorithm to enhance the image contrast, using an image segmentation method to segment cement particles and hydration products in the enhanced image; extracting significant cement particle microstructures from the segmentation results, and processing them using morphological operations; obtaining an image containing only significant cement particle phases by retaining cement particles whose areas reach a set threshold in the image; Affine transformation is performed on the CT image and BSE image after significant phase extraction, including: taking the BSE image after significant phase extraction as a fixed image, taking the coordinate space of the CT image as a standard transformation space, and constructing a two-dimensional plane area with the same size as the BSE image based on the transformation parameters; transforming the two-dimensional plane area to obtain slices along a certain direction in the CT image, and taking the CT image slices as floating images matched with the BSE image; using the progressive competitive pyramid particle swarm optimization algorithm to iteratively optimize the transformation parameters to obtain the optimal transformation parameters to determine the CT image slice with the highest fitness with the BSE image; The method uses a progressive competition pyramid particle swarm optimization algorithm to iteratively optimize the transformation parameters, including: dividing the optimization process of the transformation parameters into two stages, the first stage adopts a particle competition strategy, pairs particles in pairs, and divides them into winners and losers according to fitness, the losers learn from the winners, and the winners directly enter the next generation; the second stage divides the particles into stratified populations according to fitness, and the population adopts a pyramid hierarchical topological structure; the particles in each layer of the population compete with each other, the winners learn from the individual best in history and the particles in the previous layer of the population, and the losers learn from the winners of the same layer of the population.

2. A cross-modal image registration method for cement hydration microstructure analysis according to claim 1, characterized in that: The CLAHE contrast enhancement algorithm is used to enhance the image contrast, and the K-MEANS clustering method is used to cluster and classify the image grayscale to segment the cement particles and hydration products in the image.

3. A cross-modal image registration method for cement hydration microstructure analysis according to claim 1, characterized in that: Calculating the area of ​​cement particles in the image includes: extracting a set of contour points of cement particles in the image by a boundary tracking algorithm; and calculating the area of ​​cement particles by using a polygon area calculation formula based on the set of contour points.

4. A cross-modal image registration method for cement hydration microstructure analysis according to claim 1, characterized in that: BSE images were obtained using a scanning electron microscope, and three-dimensional CT images were obtained by micro-CT scanning.

5. The cross-modal image registration method for cement hydration microstructure analysis according to claim 1, characterized in that: The affine transformation includes translation transformation and rotation transformation, and the transformation order is to first perform translation transformation and then perform rotation transformation around the X, Y, and Z coordinate axes; The translation parameter is the displacement of the two-dimensional plane area relative to the origin of the CT image; the rotation parameter is the rotation angle of the three coordinate axes, and the rotation center is the coordinate corresponding to the center position of the two-dimensional plane area.

6. A cross-modal image registration method for cement hydration microstructure analysis according to claim 1, characterized in that: The fitness between the floating image and the BSE image is calculated based on the DICE coefficient.

7. A cross-modal image registration system for cement hydration microstructure analysis, characterized in that: include: The image acquisition and preprocessing module is configured to: acquire cement hydration microstructure images, including CT images and BSE images; The image significant phase extraction module is configured to: extract significant phases from the acquired image, including: using a contrast enhancement algorithm to enhance the image contrast, using an image segmentation method to segment cement particles and hydration products in the enhanced image; extracting significant cement particle microstructures of the image from the segmentation results, and processing them using morphological operations; obtaining an image containing only significant cement particle phases by retaining cement particles whose areas reach a set threshold in the image; The image registration module is configured to: perform affine transformation on the CT image and the BSE image after significant phase extraction, including: taking the BSE image after significant phase extraction as a fixed image, taking the coordinate space of the CT image as a standard transformation space, and constructing a two-dimensional plane area with the same size as the BSE image based on the transformation parameters; transforming the two-dimensional plane area to obtain slices along a certain direction in the CT image, and taking the CT image slices as floating images matched with the BSE image; iteratively optimizing the transformation parameters using a progressive competitive pyramid particle swarm optimization algorithm to obtain the optimal transformation parameters to determine the CT image slice with the highest fitness with the BSE image; The method uses a progressive competition pyramid particle swarm optimization algorithm to iteratively optimize the transformation parameters, including: dividing the optimization process of the transformation parameters into two stages, the first stage adopts a particle competition strategy, pairs particles in pairs, and divides them into winners and losers according to fitness, the losers learn from the winners, and the winners directly enter the next generation; the second stage divides the particles into stratified populations according to fitness, and the population adopts a pyramid hierarchical topological structure; the particles in each layer of the population compete with each other, the winners learn from the individual best in history and the particles in the previous layer of the population, and the losers learn from the winners of the same layer of the population.

8. An electronic device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the cross-modal image registration method for cement hydration microstructure analysis as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instruction is executed by a processor, the steps of the cross-modal image registration method for cement hydration microstructure analysis described in any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the cross-modal image registration method for cement hydration microstructure analysis described in any one of claims 1 to 6 are implemented.

Images