CN-122023909-A - Method for analyzing structural morphology of catalyst based on computer vision

CN122023909ACN 122023909 ACN122023909 ACN 122023909ACN-122023909-A

Abstract

The invention discloses a catalyst structure morphology analysis method based on computer vision, in particular to the field of computer vision, which is used for solving the problem that the segmentation precision and the pore space communication structure stability are difficult to ensure simultaneously under the condition of complex pore morphology in the existing catalyst three-dimensional structure analysis process. The method comprises the steps of constructing a multidimensional feature vector for a three-dimensional gray image body of a catalyst, carrying out probability segmentation on a pore phase and a solid phase, introducing topological persistence constraint in the segmentation process, carrying out strengthening and retaining on a key pore structure which exists stably, carrying out structural correction on a segmentation result by combining a morphological processing mode of direction perception, further improving continuity and consistency of a pore channel, and finally extracting a skeleton network structure of the pore space through three-dimensional center axis transformation to realize effective expression of a pore communication relation and a channel form, thereby providing a reliable technical means for analysis and optimization of structural performance of the catalyst.

Inventors

WANG XINBO
LI XIAOJU
XIA YANFANG
LIU YAZHOU
CHU CHAO
CHEN YINGFENG

Assignees

山东大学
山东长泽新材料科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (9)

1. The method for analyzing the structural morphology of the catalyst based on computer vision is characterized by comprising the following steps of: S1, generating a multidimensional feature vector corresponding to each voxel according to input three-dimensional gray scale image volume data of a catalyst; S2, constructing an energy function minimization problem through image segmentation based on the multidimensional feature vector, and solving to obtain an initial classification probability distribution map of each voxel belonging to a foreground phase or a background phase; S3, performing sub-layer threshold processing on the initial classification probability distribution map to generate a binarized slice image, and performing continuous coherent analysis on the slice image to extract the appearance level and the disappearance level of the key topological feature; S4, constructing a topological persistence constraint item according to the appearance level and the disappearance level of the key topological feature, fusing the constraint item with an energy function during image segmentation, and solving again to obtain a final classification probability distribution map; S5, determining a local main direction on the final classification probability distribution diagram according to the Herson matrix eigenvectors at each voxel position, and constructing an asymmetric structural element along the local main direction; S6, performing directional morphological corrosion operation on the final classification probability distribution diagram by adopting asymmetric structural elements, and obtaining a final binary segmentation image after iterative convergence; and S7, applying a three-dimensional medial axis transformation algorithm to the final binary segmentation image, and extracting a skeleton network diagram representing a pore space communication structure.
2. The method for analyzing structural morphology of a catalyst based on computer vision according to claim 1, wherein in S1, generating a multidimensional feature vector corresponding to each voxel specifically includes: Three-dimensional gray image data of the catalyst are obtained and voxel extraction is carried out; constructing a local neighborhood window by taking each voxel as a center, and calculating the mean value, variance and skewness of gray values of all voxels in the local neighborhood window as local gray distribution statistical moment; Calculating a second-order hessian matrix at each voxel position, solving the characteristic values of the matrix, and calculating tubular characteristic measure and sheet characteristic measure according to the relative magnitude relation among the characteristic values; calculating a gradient amplitude diagram of the three-dimensional gray-scale image body of the catalyst, identifying local maximum points in the gradient amplitude diagram as image gradient peaks, and calculating the three-dimensional Euclidean distance from each voxel to the nearest image gradient peak; And linearly splicing the local gray distribution statistical moment, the tubular and sheet feature measure and the three-dimensional Euclidean distance to form a multidimensional feature vector corresponding to each voxel.
3. The method for analyzing the structural morphology of the catalyst based on computer vision according to claim 1, wherein in S2, constructing an energy function minimization problem by image segmentation based on the multidimensional feature vector, and solving to obtain an initial classification probability distribution map of each voxel belonging to a foreground phase or a background phase specifically comprises: respectively calculating single-point potential energy of each voxel belonging to a foreground phase and a background phase according to the multidimensional feature vector; Constructing edges between all adjacent voxels, and calculating the potential energy of a smoothing item of each edge according to the difference of initial multidimensional feature vectors between the adjacent voxels; All voxels are used as nodes, single-point potential energy and smooth item potential energy are assigned to corresponding nodes and edges to construct a weighted graph structure; and solving the minimum cut of the energy function by applying a maximum flow minimum cut algorithm on the graph structure, marking the nodes according to the minimum cut result, converting the marking result into a probability value of each voxel belonging to a foreground phase, and forming the initial classification probability distribution map.
4. The method for analyzing the structural morphology of the catalyst based on computer vision according to claim 3, wherein the energy function is a Markov random field energy function constructed based on a graph-cut algorithm, and the data item formed by the single-point potential energy and the smooth item formed by the smooth item potential energy are constructed in a linear superposition mode.
5. The method for analyzing the structural morphology of the catalyst based on computer vision according to claim 1, wherein in S3, extracting the appearance level and the disappearance level of the key topological feature specifically includes: setting a sub-layer threshold sequence to divide an initial classification probability distribution map, generating a corresponding binary image sequence, marking three-dimensional connected components of each binary image in the binary image sequence, and identifying all connected pore areas in the binary image sequence; tracking the occurrence, combination and disappearance process of each connected pore area along the sequence direction of the sub-layer threshold sequence along with the change of the threshold value; Recording a corresponding threshold value as a presentation level of the topological feature when one connected pore region appears for the first time, recording as a merging event when more than two connected pore regions are merged, and recording a corresponding threshold value as a disappearance level of the topological feature when one connected pore region completely disappears; and summarizing all connected pore areas which are not covered by the merging event, and taking the corresponding appearance level and disappearance level as the persistence data of the key topological feature.
6. The method for analyzing the structural morphology of the catalyst based on computer vision according to claim 1, wherein in the step S4, according to the appearance level and the disappearance level of the key topological feature, a topological persistence constraint term is constructed, the constraint term is fused with an energy function when the image is segmented, and the step of re-solving to obtain a final classification probability distribution map specifically comprises: Calculating the persistence value of each feature according to the appearance level and the disappearance level of the key topological feature, and screening out the feature with the persistence value exceeding a preset persistence threshold value as the key feature; Extracting continuous connected areas between the appearance level and the disappearance level of the key features, and recording all voxel coordinates forming the connected areas as a key voxel set; constructing a penalty function based on voxel attribution as a topology persistence constraint item, wherein the penalty function applies a constant penalty value when voxels belonging to a key voxel set are marked as background phases; And fusing the topological persistence constraint item with the energy function, solving again by applying a maximum flow minimum cut algorithm, and generating a final classification probability distribution diagram according to a solving result.
7. The method for analyzing the structural morphology of the catalyst based on computer vision according to claim 1, wherein in S5, determining a local principal direction according to the hessian matrix eigenvectors at each voxel position on the final classification probability distribution map, and constructing asymmetric structural elements along the local principal direction specifically comprises: calculating a second-order hessian matrix at each voxel position on the final classification probability distribution diagram, carrying out feature decomposition on the matrix to obtain a feature value, screening a feature vector with the minimum absolute value of the feature value, and determining the direction indicated by the vector as a local main direction; defining an ellipsoid prototype under a three-dimensional orthogonal coordinate system by taking a local main direction as a long axis direction, carrying out fixed scale compression on the ellipsoid prototype on a cross section perpendicular to the local main direction, and constructing an asymmetric three-dimensional structure element extending along the main direction, wherein the center of the asymmetric three-dimensional structure element is aligned with the current voxel position.
8. The method for analyzing the structural morphology of the catalyst based on computer vision according to claim 1, wherein in S6, performing a directional morphological erosion operation on the final classification probability distribution map by using an asymmetric structural element, and obtaining a final binary segmentation image after iterative convergence specifically comprises: The final classification probability distribution map is used as an input image, and the asymmetric structural elements are used as operation templates; Sliding an asymmetric structural element on an input image, calculating the probability minimum value of all voxels in a space covered by the structural element at each position, and replacing the probability value of a central voxel with the minimum value to finish one-time guiding corrosion operation; repeatedly executing the directional etching operation by taking the image obtained by the etching operation as a new input image until the variation of all voxel probability values between images generated by two continuous etching operations is smaller than a preset tolerance threshold value, and judging that iteration converges; and carrying out binarization judgment on the probability value of each voxel in the image after iteration convergence, and generating a final binary segmentation image.
9. The method for analyzing the structural morphology of the catalyst based on computer vision according to claim 1, wherein in S7, applying a three-dimensional medial axis transformation algorithm to the final binary segmentation image, extracting a skeleton network map representing the pore space communication structure specifically comprises: And constructing a connection graph according to the space adjacent relation and the distance field value by taking the initial skeleton point as a node, solving a minimum spanning tree in the connection graph, and outputting nodes and edges according to the minimum spanning tree to form a skeleton network graph.

Description

Method for analyzing structural morphology of catalyst based on computer vision Technical Field The invention relates to the technical field of computer image processing, in particular to a catalyst structure morphology analysis method based on computer vision. Background In the processes of catalyst research and development, performance evaluation and failure mechanism analysis, the spatial morphology and communication characteristics of the internal pore structure of the catalyst are important factors influencing the diffusion efficiency of reactants, the utilization rate of active sites and the overall catalytic performance. With the development of three-dimensional imaging technology, three-dimensional gray images with micron-scale and nano-scale resolutions can more completely reflect the complex pore channel distribution and structure morphology inside the catalyst, and a data basis is provided for refined structure analysis. In practical application, the three-dimensional image of the catalyst generally has the problems of uneven gray distribution, obvious structural scale difference, fuzzy pore boundaries, noise interference and the like, so that the pore phase and the entity phase are difficult to accurately distinguish only by relying on manual experience or a simple threshold segmentation mode, and especially in the areas with long and thin pore channels, mutual interweaving or weaker local connectivity, the conditions of structural fracture, false separation or false communication are easy to occur. The traditional image segmentation or morphological processing-based method focuses on local gray scale or geometric characteristics in multiple ways, lacks of characterization on the stability of the overall communication structure of the pore space, and is difficult to maintain global structural consistency while guaranteeing local fine segmentation. In addition, the segmentation result is usually presented in a voxel or volume form, which is not beneficial to intuitively analyzing and quantitatively characterizing the communication relation, the channel trend and the topological structure of the pore network, thereby limiting the practical value of the segmentation result in application scenes such as catalyst structure optimization, performance contrast analysis, multi-scale transmission modeling and the like. Therefore, a method for analyzing the structural morphology of a catalyst, which can combine three-dimensional image characteristics, structural stability constraints and spatial connectivity analysis, is needed to realize reliable identification and effective expression of complex pore structures. Disclosure of Invention In order to overcome the above-mentioned drawbacks of the prior art, embodiments of the present invention provide a method for analyzing structural morphology of a catalyst based on computer vision to solve the above-mentioned problems in the prior art. In order to achieve the above purpose, the present invention provides the following technical solutions: a method for analyzing the structural morphology of a catalyst based on computer vision, comprising the following steps: S1, generating a multidimensional feature vector corresponding to each voxel according to input three-dimensional gray scale image volume data of a catalyst; S2, constructing an energy function minimization problem through image segmentation based on the multidimensional feature vector, and solving to obtain an initial classification probability distribution map of each voxel belonging to a foreground phase or a background phase; S3, performing sub-layer threshold processing on the initial classification probability distribution map to generate a binarized slice image, and performing continuous coherent analysis on the slice image to extract the appearance level and the disappearance level of the key topological feature; S4, constructing a topological persistence constraint item according to the appearance level and the disappearance level of the key topological feature, fusing the constraint item with an energy function during image segmentation, and solving again to obtain a final classification probability distribution map; S5, determining a local main direction on the final classification probability distribution diagram according to the Herson matrix eigenvectors at each voxel position, and constructing an asymmetric structural element along the local main direction; S6, performing directional morphological corrosion operation on the final classification probability distribution diagram by adopting asymmetric structural elements, and obtaining a final binary segmentation image after iterative convergence; and S7, applying a three-dimensional medial axis transformation algorithm to the final binary segmentation image, and extracting a skeleton network diagram representing a pore space communication structure. As a further aspect of the present invention, in S1, generating the multidimensional featu