Design And Implementation Of Streamline Selection System Based On Unsupervised Learning

In the current field of science and technology,the development of fluid dynamics and flow field research has become the basis of many cutting-edge technologies,such as aerospace technology,ship design,and medical fields.As a tool of flow field visualization,streamline can help people understand the movement law and characteristics of fluid in flow field more clearly.In order to describe the characteristics of the flow field comprehensively,a large number of streamlines are often generated,but this also brings problems such as mutual occlusion between streamlines and messy display.To tackle this problem,researchers have suggested using the technique of clustering analysis for streamline selection.By clustering the streamlines with the same characteristics into a class,a small number of streamlines are selected to achieve the purpose of streamline simplification.However,the current clustering analysis is based on a single feature view,which leads to the simplification of streamlines more or less missing part of the flow field characteristics.Therefore,this paper implements the clustering screening of streamlines from the perspective of multi-view clustering.The specific work is as follows:(1)A variety of streamline feature extraction methods are implemented and improved.In order to describe the streamline comprehensively,this paper extracts the streamline feature from two angles of spatial position and geometric shape,implements a variety of streamline feature extraction methods,and improves the method based on segmentation.(2)Two streamline multi-view spectral clustering algorithms are implemented.In order to make the clustering effect more accurate,this paper applies a multi-view spectral clustering algorithm based on co-training and co-regularization to cluster the extracted streamline feature views.By imposing constraints on the spatial position and geometric shape of the streamline,the division of the streamline is more accurate and the clustering effect is better.(3)A multi-view hierarchical ensemble clustering framework for streamline clustering is constructed.The framework adopts the idea of ensemble clustering,improves the effectiveness of information mining from streamline views by applying different clustering algorithms to different views,and improves the accuracy and stability of clustering algorithms by fusing multi-level clustering results.(4)Develop a streamline selection system integrating unsupervised learning methods.In this software system,interactive streamline browsing and clustering analysis are realized,and good analysis results are obtained.