Three-dimensional flow field feature tracking method based on fusion attribute distribution similarity

By integrating the similarity of attribute distribution, the attribute vectors of flow field features are extracted and principal component analysis, cluster analysis and curve similarity calculation are carried out, which solves the problem of tracking interrupts and calculation overhead in three-dimensional time-varying flow field feature tracking. Efficient feature matching and event detection are achieved.

CN118916840AInactive Publication Date: 2024-11-08SICHUAN UNIVERSITY OF SCIENCE AND ENGINEERING
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411405916.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-10
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing three-dimensional time-varying flow field feature tracking methods are prone to tracking interrupt problems when dealing with fast moving or densely distributed features. The methods based on attribute consistency and spatial consistency require a large amount of calculation, resulting in increased calculation overhead and inability to effectively detect. Events in the process of character development.

Method used

Using a three-dimensional flow field feature tracking method based on the similarity of the distribution of fusion attributes, the attribute matrix is ​​constructed by extracting the attribute vectors of the vortex features, and principal component analysis is performed to reduce the dimensions. Cluster analysis obtains the distribution curve of the feature data, calculates the curve similarity, and generates feature similarity. degree matrix, realize feature matching and event analysis.

Benefits of technology

Effectively combining the physical information of the vortex with spatial information, improves matching efficiency, reduces the amount of calculation, and can accurately detect events in the development of features, and is suitable for large-scale complex flow field feature tracking.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118916840A_ABST
    Figure CN118916840A_ABST
Patent Text Reader

Abstract

The invention discloses a three-dimensional flow field feature tracking method based on fusion attribute distribution similarity, and relates to the technical field of flow field visualization, and the method comprises the steps: extracting vortex features, carrying out the attribute calculation based on the vortex features to obtain an attribute vector, and constructing an attribute matrix for describing the vortex features according to the attribute vector; performing dimension reduction on the attribute matrix by adopting principal component analysis to obtain a fusion matrix; analyzing all fusion matrixes of adjacent time steps based on a clustering method to obtain a feature data distribution curve; calculating the similarity of the characteristic data distribution curve according to a curve similarity calculation formula to obtain a characteristic similarity matrix; and performing feature matching on the features in the current adjacent time steps according to the feature similarity matrix to obtain a feature event, and analyzing an unsteady flow field according to the feature event to complete three-dimensional flow field feature tracking. According to the method, the physical information and the space information of the vortex are effectively combined, and the matching efficiency is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of flow field visualization, and in particular relates to a three-dimensional flow field feature tracking method based on fusion attribute distribution similarity. Background Art

[0002] The essence of vortex feature tracking is to find the feature correspondence between two consecutive time steps based on dense time sampling. However, in actual operation, the time sampling of flow field data may not be dense enough, resulting in tracking interruption problems. In order to solve this problem, people have developed various feature tracking technologies focusing on time-varying flow fields. At present, the methods for tracking features in the field of flow field visualization are mainly divided into two categories: attribute consistency-based methods and spatial consistency-based methods. Haak et al. and Kalivas et al. designed a flow field feature tracking algorithm based on spatial consistency. Common matching methods based on spatial consistency include volume overlap, position, maximum cross-correlation criterion, minimum affine transformation matrix, etc. Silver and Wang designed a method based on volume overlap to solve the efficiency problem of processing a large number of time-varying data sets. Muelder C designed a tracking method based on feature position, which predicts the feature position of adjacent time steps according to the position curve of the tracked feature, sets a threshold for the deviation between the predicted value and the true value for judgment, and corrects the predicted value to continue the prediction. A. Schnorr used the graph optimization method to transform the continuous time step feature matching problem into a graph matching problem and proposed a two-stage feature tracking method, but it is not suitable for three-dimensional time-varying flow fields. Although experiments have shown that the above methods can effectively and quickly match and track features, the tracking method is prone to tracking interruption when dealing with fast-moving features or features from dense distribution. To solve the problem of tracking interruption, P. Pan proposed a feature similarity measurement method based on spatial distribution optimization. This method uses the overlap of feature volumes combined with the spatial distribution information of features to match features, but the addition of spatial distribution information increases the computational overhead and cost. The attribute consistency-based method is different from only considering the spatial information of features in space. It usually expresses the attributes of features and calculates the similarity measure of attributes between adjacent feature steps to match features. In the field of computer vision, models based on physical properties are often used for tracking and three-dimensional representation. However, these methods usually cannot classify events and cannot cope well with changes in topological structures. But similar to the ideas of these methods, in the field of flow field visualization, Ji G proposed a method based on feature attributes, such as volume, weight, centroid, and second-order moment, to match features. If the difference between the centroid, weight, and volume of an object at two consecutive time points is within a specified small threshold range, the two features match each other. Abdel Kareem W proposed an algorithm to track the characteristics of tubular vortices in three-dimensional time-varying flow fields. The tubular vortices in three-dimensional flow fields are described and the characteristics of adjacent time steps are matched. Tzeng et al. proposed a new method based on machine learning to solve the problem of extracting and visualizing 4D features in large amounts of data.The attribute consistency method tracks features based on physical attributes and can solve the tracking problem to a certain extent, but it often requires the calculation of a large number of physical attributes, which increases the cost of feature tracking and reduces the tracking efficiency. Cluster analysis is one of the most important techniques in unsupervised machine learning and has been widely used in flow field feature extraction, such as clustering based on physical feature parameters, clustering based on streamlines or traces, clustering based on time activity (TAC) curves, and other vortex extraction methods (Integral Curve Clustering). Kuhn et al. proposed a flow field feature recognition technology based on density clustering and successfully applied it to the vortex extraction of three-dimensional airfoil flow data. Hadjighasem et al., Meuschke et al., and El et al. directly applied different clustering methods based on traces to realize the identification of vortex features. Shi et al. conducted a comprehensive comparative study on flow field visual analysis technology based on clustering of integral curves such as streamlines and traces, providing guidance for researchers to choose appropriate clustering methods and similarity metrics. Padmesh et al. proposed a vortex extraction and visualization method based on K-menas clustering. Based on the integral curve of the fusion physical characteristic parameters, Nguyen et al. proposed a visual analysis framework for unsteady flow fields based on spatiotemporal hierarchical clustering. The above algorithm has achieved good vortex recognition results. Feature tracking requires monitoring the similarity relationship between multiple features in the time dimension, while the aforementioned method is only applicable to extracting vortices in a single time step and cannot detect characteristic events in the feature development process. At present, there are few unsupervised methods for tracking three-dimensional time-varying flow field features. The existing attribute consistency-based methods and spatial consistency-based methods both match the relationship between features by calculating the similarity measure between multi-time step features, which requires a lot of calculations. Due to the complexity of the vortex phenomenon, as the data scale increases, the computational overhead will also increase significantly, and the method based on spatial consistency lacks the description of the physical characteristics of the features. Therefore, it is very necessary to design an intelligent tracking method for vortex features that integrates flow field characteristics. Summary of the invention

[0003] To solve the above technical problems, the present invention proposes a three-dimensional flow field feature tracking method based on the similarity of fused attribute distribution, which effectively combines the physical information of the vortex with the spatial information, improves the matching efficiency, and provides a new idea for tracking large-scale complex flow field features.

[0004] To achieve the above object, the present invention provides a three-dimensional flow field feature tracking method based on fusion attribute distribution similarity, comprising: Extracting vortex features, performing attribute calculation based on the vortex features to obtain an attribute vector, and constructing an attribute matrix describing the vortex features according to the attribute vector; Using principal component analysis to reduce the dimension of the attribute matrix to obtain a fusion matrix; All fusion matrices of adjacent time steps are analyzed based on the clustering method to obtain the characteristic data distribution curve; According to the curve similarity calculation formula, the similarity of the characteristic data distribution curve is calculated to obtain the characteristic similarity matrix; According to the feature similarity matrix, feature matching is performed on the features in the current adjacent time step to obtain feature events, and the unsteady flow field is analyzed according to the feature events to complete three-dimensional flow field feature tracking.

[0005] Optionally, a method for obtaining an attribute vector by performing attribute calculation based on the vortex feature is:

[0006] Among them, A1 is the Q value calculated based on the Q criterion; A2 is the vorticity size; A3 is the local shear rate; A4 is F-norm of a matrix ; A5 is Point coordinates; is the attribute vector χ at any point; is the Q value calculated based on the Q criterion; is the tension tensor matrix at point x; For point The vorticity tensor at ; is the vorticity with velocity vector v in the flow field; is the vorticity; is the velocity vector; is the gradient symbol.

[0007] Optionally, a method for constructing a property matrix describing vortex characteristics according to the property vector is: =

[0008] Among them, m is the current vortex area containing m points.

[0009] Optionally, principal component analysis is used to reduce the dimension of the attribute matrix to obtain a fusion matrix including: Standardizing the attribute matrix to obtain a standardized matrix; Calculating the covariance matrix of the standardized matrix, performing eigenvalue decomposition on the covariance matrix, and calculating the eigenvalues ​​and corresponding eigenvectors of the covariance matrix; Constructing a projection matrix based on the eigenvalues ​​and corresponding eigenvectors of the covariance matrix; The normalized matrix is ​​projected into a new low-dimensional space according to the projection matrix to obtain a fusion matrix.

[0010] Optionally, all fusion matrices of adjacent time steps are analyzed based on a clustering method to obtain a characteristic data distribution curve:

[0011] Among them, t is the sampling time of the time series flow field, k and j are the numbers of the vortices in the flow field at time t and time t+1 respectively. There are n vortices at time t and m vortices at time t+1. i represents the number of the cluster in the clustering result. There are n+m clusters in total. is the distribution frequency of the kth feature in the flow field at time t in cluster i, that is, the value of the curve Г at point i; is the distribution frequency of feature k in cluster i at time t; is the distribution frequency of feature j in cluster i at time t+1; is the distribution frequency of feature j in cluster i at time t+1.

[0012] Optionally, calculating the similarity of the characteristic data distribution curve according to the curve similarity calculation formula to obtain the characteristic similarity matrix includes: The distance between the two curves is obtained according to the similarity measurement, and the feature similarity matrix is ​​constructed according to the similarity between the two distribution curves. The distance between the two curves is calculated as follows:

[0013] in, is the distance between two curves; is the characteristic distribution curve numbered j at the previous moment; is the characteristic distribution curve numbered k at the next moment; is the distribution curve of feature j at the previous moment at point i; The distribution curve of feature k at the next moment takes the value at i; is the number of clusters in the clustering result; j and k are the feature numbers of the previous moment and the next moment respectively; i is the cluster number.

[0014] Optionally, the feature similarity matrix includes a matrix row number representing the number of the time step feature region, and a matrix column number representing the number of the time step feature region; the matrix elements are 0 or 1, 0 indicates that the similarity between the two feature regions is less than the minimum similarity threshold, and the two feature regions are not similar; 1 indicates that the similarity between the two feature regions is greater than the minimum similarity threshold, and the two feature regions are similar.

[0015] Optionally, the characteristic events include continuation events, split events, merge events, generation events, and dissipation events; among which 1 to 1 is a continuation event, 1 to many is a split event, many to 1 is a merge event, 0 to 1 is a generation event, and 1 to 0 is a dissipation event.

[0016] Technical effect of the invention: The invention discloses a three-dimensional flow field feature tracking method based on fusion attribute distribution similarity, obtains data distribution characteristics through cluster analysis, calculates distribution curve similarity measurement to match feature relationships; effectively constructs an attribute matrix with multiple physical attributes of flow field characteristics and fuses them, and describes the physical characteristics of the vortex phenomenon in the flow field; calculates distribution curve similarity instead of feature similarity measurement calculation in existing methods, thereby reducing the amount of calculation and improving feature matching efficiency while ensuring accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A schematic flow chart of a three-dimensional flow field feature tracking method based on fusion attribute distribution similarity according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the distribution frequency of fusion vectors in clusters according to an embodiment of the present invention; Figure 3 A distribution curve diagram for describing the distribution of the vortex feature fusion matrix according to an embodiment of the present invention; Figure 4 It is a heat map of similarity between features of adjacent time steps in an embodiment of the present invention; Figure 5 A directed acyclic graph that describes the relationship between features in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0019] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0020] like Figure 1 As shown, this embodiment provides a three-dimensional flow field feature tracking method based on fusion attribute distribution similarity, including: Extract vortex characteristics, calculate attributes based on the vortex characteristics to obtain attribute vectors, and construct an attribute matrix describing the vortex characteristics based on the attribute vectors; The principal component analysis is used to reduce the dimension of the attribute matrix and obtain the fusion matrix; All fusion matrices of adjacent time steps are analyzed based on the clustering method to obtain the characteristic data distribution curve; According to the curve similarity calculation formula, the similarity of the characteristic data distribution curve is calculated to obtain the characteristic similarity matrix; According to the feature similarity matrix, the features in the current adjacent time step are matched to obtain feature events, and the unsteady flow field is analyzed according to the feature events to complete the three-dimensional flow field feature tracking.

[0021] Specifically, this embodiment obtains data distribution characteristics through cluster analysis, calculates distribution curve similarity measurement to match feature relationships; effectively constructs a property matrix with multiple physical attributes of flow field characteristics and fuses them, and describes the physical characteristics of the vortex phenomenon in the flow field; calculates distribution curve similarity instead of feature similarity measurement calculation in existing methods, thereby reducing the amount of calculation and improving feature matching efficiency while ensuring accuracy.

[0022] Further, such as Figure 2 As shown in the figure, the distribution frequency of the vectors in each vortex matrix in each cluster is described. The horizontal axis is the feature number, such as t1_f1 represents feature No. 1 at time t1, t1_f2 represents feature No. 2 at time t1; t2_f1 represents feature No. 1 at time t2, t2_f2 represents feature No. 2 at time t2, and t2_f3 represents feature No. 3 at time t2; the vertical axis is the distribution frequency, and the proportion of different gray levels in the bar graph represents the distribution frequency of the current feature in the cluster represented by the gray level. For example, the bar graph with the horizontal axis t1_f1 represents that feature f1 at time t1 is mainly distributed in clusters No. 1 and No. 5. The distribution frequencies are 94% and 6% respectively; t1_f2 is mainly distributed in clusters numbered 1, 2, 3 and 4, with distribution frequencies of 0.3%, 48.9%, 4.6% and 46.2% respectively; t2_f1 is mainly distributed in clusters numbered 1 and 5, with distribution frequencies of 92.7% and 7.3% respectively; t2_f2 is mainly distributed in clusters numbered 1 and 5, with distribution frequencies of 96.7% and 3.3% respectively; t2_f3 is mainly distributed in clusters numbered 1, 2, 3 and 4, with distribution frequencies of 0.3%, 50%, 5.4% and 44.3% respectively. The method for obtaining attribute vectors by calculating attributes based on vortex characteristics is:

[0023] Among them, A1 is the Q value calculated based on the Q criterion; A2 is the vorticity size; A3 is the local shear rate; A4 is F-norm of a matrix ; A5 is Point coordinates; is the attribute vector χ at any point; is the Q value calculated based on the Q criterion; is the tension tensor matrix at point x; For point The vorticity tensor at ; is the vorticity with velocity vector v in the flow field; is the vorticity; is the velocity vector; is the gradient symbol.

[0024] Specifically, take the time series flow field Moment flow field Any point within , whose velocity vector is ,That The matrix is ​​represented as ,Right now .in is the point tension tensor matrix (straintensor), For point The vorticity tensor at .

[0025] Selected attributes are widely used in flow analysis to describe different physical characteristics of the flow field, e.g. and vorticity are often used to describe the rotational characteristics of vortex features, while local shear rate and It is often used to study the divergence and separation behavior of local flow fields.

[0026] Furthermore, the attribute matrix describing the vortex characteristics is constructed based on the attribute vector After obtaining the attribute vector of each vortex feature, simply concatenate the attribute vectors to obtain the original matrix = , where m means that the current vortex area contains m points.

[0027] Combining spatial attributes with physical attributes, a relatively complete attribute matrix describing vortex characteristics is constructed.

[0028] Furthermore, principal component analysis is used to reduce the dimension of the attribute matrix, and the fusion matrix obtained includes: Standardize the attribute matrix to obtain a standardized matrix; Calculate the covariance matrix of the standardized matrix, perform eigenvalue decomposition on the covariance matrix, and calculate the eigenvalues ​​and corresponding eigenvectors of the covariance matrix; Construct a projection matrix based on the eigenvalues ​​of the covariance matrix and the corresponding eigenvectors; The normalized matrix is ​​projected into a new low-dimensional space according to the projection matrix to obtain a fusion matrix.

[0029] Specifically, since there is an essential connection between A1, A2, ..., A4, directly using Clustering the matrix will generate noise, which will affect the clustering results. In addition, this redundant information will increase the computational cost and reduce the efficiency of the algorithm. It is necessary to simplify the attributes involved in tracking while retaining the characteristics of the attribute vector to the greatest extent. Therefore, PCA principal component analysis is used to reduce the dimension of the attribute matrix to obtain the fusion matrix The method comprises the following steps: The matrix is ​​standardized by column, that is, each attribute of the vortex is standardized, the purpose is to make the mean of each attribute 0 and the standard deviation 1, to avoid the data differences between different attributes affecting the results. For each attribute value Xij in the matrix X, calculate the standardized value:

[0030] in, For each attribute value in X; is the mean of the attribute values ​​in column j; is the variance of the attribute values ​​in column j.

[0031] Calculate the covariance matrix of the standardized matrix , , is the standardized attribute matrix, for The transposed matrix of for The data matrix is ​​obtained by performing eigenvalue decomposition on the covariance matrix C and calculating the eigenvalues ​​and corresponding eigenvectors of the covariance matrix. The eigenvalue reflects the variance in the direction of each eigenvector. Each eigenvalue corresponds to an eigenvector. In the experiment, the contribution of the selected components to the description of the vortex characteristics is statistically analyzed when k_pca (k=1,2,3,4,5) eigenvectors are taken respectively. When k_pca=5, the contribution value shows an obvious upward trend, while when k_pca=5, the contribution value growth trend decreases significantly. According to the elbow rule, 4 eigenvalues ​​and their corresponding eigenvectors are selected, and their contribution is 88%, as the principal components to form the projection matrix P, and the size of P is 7×k_pca. The projection matrix P is used to project the standardized original data X into a new low-dimensional space. , is a matrix of size m×k_pca. It is the target matrix, that is, the main component attribute matrix describing the characteristic area of ​​the vortex. The original matrix is ​​simplified to reduce the amount of calculation for subsequent calculations.

[0032] Furthermore, given any two points x and y in the flow field, after attribute fusion, their fusion vector is , the closer the distance between vectors, the more similar the vectors are. One of the keys to the k-means clustering analysis method is to determine the value of the parameter k_cluster, that is, the number of clusters obtained in the end. Through the analysis of the flow field vortex, it is believed that if the number of features at time t is n and the number of vortices at time t+1 is m, no matter what events occur from t to t+1, the maximum number of completely independent vortices that may be obtained is n+m, that is, all vortices at time t are not related to the vortices at time t+1. is the number of cluster centers, each fused attribute vector is a sample, and the rows of all principal component attribute matrices of adjacent times are clustered. Vectors with close distances are divided into the same cluster through k-means clustering. Figure 3 As shown, different curves represent different vortex characteristics. Each curve describes the distribution frequency of the vectors in each vortex matrix in each cluster. The horizontal axis is the cluster number, and the vertical axis is the distribution frequency. The value of the curve at the corresponding horizontal axis represents the distribution frequency of the characteristic described by the current curve in the cluster. The curve of t1_f1 indicates that the distribution frequencies of feature f1 at time t1 in clusters numbered 1 and 5 are respectively 0.94, 0, 0, 0, 0.06; the distribution frequencies of t1_f2 in clusters are respectively 0.003, 0.489, 0.046 and 0.462, 0; the distribution frequencies of t2_f1 in clusters are respectively 0.927, 0, 0, 0, 0.073; the distribution frequencies of t2_f2 in clusters are respectively 0.967, 0, 0, 0, and 0.033; the distribution frequencies of t2_f3 in clusters are respectively 0.003, 0.5, 0.054 and 0.443, 0; the distribution frequencies of vectors in each vortex matrix in each cluster are obtained. To describe the distribution characteristics of the vectors in the fusion matrix, the following distribution curve is defined When the clustering results converge, the distribution frequency of each vortex in each cluster is counted, and the vortex distribution curve numbered k at time t can be obtained.

[0033]

[0034] Among them, t is the sampling time of the time series flow field, k is the number of the vortex in the flow field at time t, is the distribution frequency of the kth feature in the flow field at time t in cluster i.

[0035] In order to simplify the symbols, the characteristic frequency distribution curve in the adjacent time flow field is expressed as:

[0036] Among them, t is the sampling time of the time series flow field, k and j are the numbers of the vortices in the flow field at time t and time t+1 respectively. There are n vortices at time t and m vortices at time t+1. i represents the number of the cluster in the clustering result. There are n+m clusters in total. is the distribution frequency of the kth feature in the flow field at time t in cluster i, that is, the value of the curve Г at point i; is the distribution frequency of feature k in cluster i at time t; is the distribution frequency of feature j in cluster i at time t+1; is the distribution frequency of feature j in cluster i at time t+1.

[0037] The curve describes the distribution characteristics of the attribute data of the feature. By comparing the similarity of the curves, the similarity between vortex features can be indirectly judged.

[0038] Furthermore, according to the curve similarity calculation formula, the similarity of the characteristic data distribution curve is calculated to obtain the characteristic similarity matrix including: The distance between the two curves is obtained according to the similarity measurement, and the feature similarity matrix is ​​constructed according to the similarity between the two distribution curves.

[0039] Specifically, Figure 4 As shown in the figure, the similarity between all features at adjacent times is described. The vertical axis represents the features at time t1, such as t1_f1 and t1_f2, which represent the vortex numbered 1 and 2 at time t1 respectively; the horizontal axis represents the vortex features at time t2, and t2_f1, t2_f2, and t2_f3 represent the vortex numbered 1, 2, and 3 at time t2 respectively; the values ​​in the figure represent the similarity between the vortices represented by the corresponding horizontal and vertical axes, such as 0.01 represents the similarity between the two cross-time features t1_f2 and t2_f1 is 0.01, 0.08 represents the similarity between the features t1_f2 and t2_f2 is 0.08, and so on. In order to compare the differences between the features of two curves, traditional distance measurement methods, such as Euclidean distance and Pearson correlation coefficient, the former is more suitable for comparing the size gap between curves, while the latter is more suitable for measuring the change trend of curves, which does not meet the need to consider the change and size of curves at the same time. To address this limitation, the TAC similarity measure (TSM) method proposed by Nguyen et al. is used to calculate the similarity of curves based on their spatial and correlation differences. The computational complexity is only linear time, so it is suitable for large-scale flow field data. TSM is defined as follows: DTSM( ) = (1 + pc Dcorr ( ))De( )

[0040] yes and The covariance of It is a curve The standard deviation of .

[0041]

[0042] in, For curve and similarity; For weighted pc , Global correlation between curves; is the distance between two curves; is the vortex characteristic distribution curve numbered j at the previous time moment; is the vortex characteristic distribution curve numbered k at the next time moment; For curve Take the value at i; For curve Take the value at i; is the total number of clusters; is the cluster number; for , Global correlation of the two curves; For curve The standard deviation of For curve The standard deviation of .

[0043] Representation curve and The Euclidean distance of The distance between curves is taken into account, and It measures the global correlation between the two curves. and When the trends are the same, , DTSM( , ) The value of the first term is 1, and the similarity of the curves depends on the Euclidean distance between the curves. If the trend is opposite, the curve distance should be larger. The penalty factor pc is introduced to assign weights to the global correlation and spatial distance. The higher the similarity Pc value of the two curves is, the greater the weight given to the global correlation. By default, Pc=1 is set.

[0044] Using the distance between the two curves obtained by the above similarity measurement method, an n× The feature similarity matrix Sim, the row number of the matrix represents The number of the time step feature region, the column number of the matrix represents The number of the feature region at the time step. The matrix elements are 0 or 1. 0 means that the similarity between the two feature regions is less than the minimum similarity threshold, that is, the two feature regions are not similar. 1 means that the similarity between the two feature regions is greater than the minimum similarity threshold, that is, the two feature regions are similar. Figure 2 Calculating the features shown will yield Figure 4 The similarity matrix shown describes the similarity between all features at adjacent times. The vertical axis represents the features at time t1, such as t1_f1 and t1_f2, which respectively represent the vortex numbered 1 and 2 at time t1; the horizontal axis represents the vortex features at time t2; the numerical values ​​in the figure represent the similarity between the vortices represented by the corresponding horizontal axis and the vertical axis.

[0045] The similarity matrix describes the similarity relationship between features. Through the similarity matrix, the correlation between features in the time dimension can be determined.

[0046] Furthermore, characteristic events include continuation events, splitting events, merging events, generation events, and dissipation events; among which 1 to 1 is a continuation event, 1 to many is a splitting event, many to 1 is a merging event, 0 to 1 is a generation event, and 1 to 0 is a dissipation event.

[0047] Specifically, the main purpose of feature tracking is to accurately describe the development and changes of features on the timeline. In the process of feature development, changes may occur, such as deformation, bifurcation, merging, etc. Event detection is to describe the development and changes of features. The types of feature change events are: continuation, birth, dissipation, split, and merge. The similarity matrix is ​​traversed to determine the relationship between features in adjacent time steps. Due to the different degrees of change in the vortex features of different unsteady flow fields, it is impossible to give a similarity threshold to judge the feature relationship. Therefore, when traversing the matrix, if the similarity between the current two features is much greater than the similarity with other features, the features are considered to match. According to the definition of feature events, there are a total of 5 events: continuation, splitting, merging, generation, and dissipation. The similarity matrix obtained by traversing the breadth-first algorithm is used to determine the events that occurred based on the feature correspondence. When the number of feature matches and the event correspondence are: 1 to 1: Continue the event; 1 to many: Split event; When there are many to one: merge events; 0 to 1: an event is generated; 1 to 0: Dissipation event.

[0048] In the development process of time-varying flow field, the characteristic events of features may span multiple time steps and have multiple changes. The results produced by the above method only get the events that occurred in two adjacent time steps. In order to show the changes of features over the entire time span, a feature tracking visualization framework is designed with reference to the directed acyclic graph (DAG). The horizontal direction represents the time development, a node represents a feature, and the edge represents the association relationship between features. According to the definition of characteristic events and the quantitative relationship between the edges between nodes, the specific type of event that occurred can be determined. Figure 5 As shown, the circle represents the vortex, and the edge with an arrow represents the association between features; the arrow points from t1_f1 to t2_f1 and t2_f2, indicating that the feature f1 at time t1 splits from time t1 to time t2, from t1_f1 to t2_f1 and t2_f2; the arrow points from t1_f2 to t2_f3, indicating that the feature f2 at time t1 undergoes a continuation event, evolving from t1_f2 to t2_f3.

[0049] The above event detection can provide insights into the characteristic events that occur during the evolution of the vortex characteristics of the unsteady flow field. The use of DAG graphs to visualize the tracking results can clearly show the development context of the vortex characteristics and provide a reference for researchers in the field.

[0050] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A three-dimensional flow field feature tracking method based on fusion attribute distribution similarity, characterized in that: include: Extracting vortex features, performing attribute calculation based on the vortex features to obtain an attribute vector, and constructing an attribute matrix describing the vortex features according to the attribute vector; Using principal component analysis to reduce the dimension of the attribute matrix to obtain a fusion matrix; All fusion matrices of adjacent time steps are analyzed based on the clustering method to obtain the characteristic data distribution curve; According to the curve similarity calculation formula, the similarity of the characteristic data distribution curve is calculated to obtain the characteristic similarity matrix; According to the feature similarity matrix, feature matching is performed on the features in the current adjacent time step to obtain feature events, and the unsteady flow field is analyzed according to the feature events to complete three-dimensional flow field feature tracking.

2. The three-dimensional flow field feature tracking method based on fusion attribute distribution similarity according to claim 1, characterized in that: The method for obtaining the attribute vector by performing attribute calculation based on the vortex characteristics is: Among them, A1 is the Q value calculated based on the Q criterion; A2 is the vorticity size; A3 is the local shear rate; A4 is F-norm of a matrix ; A5 is Point coordinates; is the attribute vector χ at any point; is the Q value calculated based on the Q criterion; is the tension tensor matrix at point x; For point The vorticity tensor at ; is the vorticity with velocity vector v in the flow field; is the vorticity; is the velocity vector; is the gradient symbol.

3. The three-dimensional flow field feature tracking method based on fusion attribute distribution similarity according to claim 1, characterized in that: The method of constructing the attribute matrix describing the vortex characteristics according to the attribute vector is: Among them, m is the current vortex area containing m points.

4. The three-dimensional flow field feature tracking method based on fusion attribute distribution similarity according to claim 1, characterized in that: The principal component analysis is used to reduce the dimension of the attribute matrix, and the fusion matrix obtained includes: Standardizing the attribute matrix to obtain a standardized matrix; Calculating the covariance matrix of the standardized matrix, performing eigenvalue decomposition on the covariance matrix, and calculating the eigenvalues ​​and corresponding eigenvectors of the covariance matrix; Constructing a projection matrix based on the eigenvalues ​​and corresponding eigenvectors of the covariance matrix; The normalized matrix is ​​projected into a new low-dimensional space according to the projection matrix to obtain a fusion matrix.

5. The three-dimensional flow field feature tracking method based on fusion attribute distribution similarity according to claim 1, characterized in that: Based on the clustering method, all fusion matrices of adjacent time steps are analyzed to obtain the characteristic data distribution curve: Among them, t is the sampling time of the time series flow field, k and j are the numbers of the vortices in the flow field at time t and time t+1 respectively. There are n vortices at time t and m vortices at time t+1. i represents the number of the cluster in the clustering result. There are n+m clusters in total. is the distribution frequency of the kth feature in the flow field at time t in cluster i, that is, the value of the curve Г at point i; is the distribution frequency of feature k in cluster i at time t; is the distribution frequency of feature j in cluster i at time t+1; is the distribution frequency of feature j in cluster i at time t+1.

6. The three-dimensional flow field feature tracking method based on fusion attribute distribution similarity according to claim 1, characterized in that: According to the curve similarity calculation formula, the similarity of the characteristic data distribution curve is calculated, and the characteristic similarity matrix is ​​obtained, including: The distance between the two curves is obtained according to the similarity measurement, and the feature similarity matrix is ​​constructed according to the similarity between the two distribution curves. The distance between the two curves is calculated as follows: in, is the distance between two curves; is the characteristic distribution curve numbered j at the previous moment; is the characteristic distribution curve numbered k at the next moment; is the distribution curve of feature j at the previous moment at point i; The distribution curve of feature k at the next moment takes the value at i; is the number of clusters in the clustering result; j and k are the feature numbers of the previous moment and the next moment respectively; i is the cluster number.

7. The three-dimensional flow field feature tracking method based on fusion attribute distribution similarity according to claim 1, characterized in that: The feature similarity matrix includes a matrix row number representing the number of the time step feature region, and a matrix column number representing the number of the time step feature region; the matrix elements are 0 or 1, 0 indicates that the similarity between the two feature regions is less than the minimum similarity threshold, and the two feature regions are not similar; 1 indicates that the similarity between the two feature regions is greater than the minimum similarity threshold, and the two feature regions are similar.

8. The three-dimensional flow field feature tracking method based on fusion attribute distribution similarity according to claim 7, characterized in that: The characteristic events include continuation events, splitting events, merging events, generating events, and dissipating events; When 1 to 1, it is a continuation event; when 1 to many, it is a splitting event; when many to 1, it is a merging event; when 0 to 1, it is a generating event; and when 1 to 0, it is a dissipating event.

Citation Information

Cited By

  • Solid waste classification method based on spectrograph

    CN119702490A

  • Positioning method and system for underwater environment

    CN121958897A