Research On The Principle And Application Of Sparse Subspace Clustering Algorithm

As an important branch of data mining,cluster analysis gradually become a hot topic in pattern recognition,image processing,and biological information analysis.The core idea of clustering analysis is to divide all data in a set of data into different categories according to the similarity between them.The process of a cluster.However,in today's society,there are a lot of high-dimensional data,such as image,multimedia,text and bioinformatics data.Due to the influence of "Curse of dimensionality",the traditional clustering algorithm has encountered bottlenecks,and the effect has been greatly reduced.Subspace clustering algorithm,as a way to solve the clustering problem of high-dimensional data,has attracted the attention of many scholars and researchers.The sparse subspace clustering algorithm based on spectral clustering has become a hot research topic due to its excellent performance.In recent years,deep neural networks have attracted much attention because of their outstanding learning ability.Compared with traditional sparse representation,deep neural networks are better at mining complex structural relationships between samples.This thesis studies the sparse representation method combined with neural networks,and proposes a sparse representation based on this sparse representation.An improved algorithm for traditional sparse subspace clustering algorithm.The main work of this thesis is as follows:1.in view of the limitation that the traditional sparse subspace clustering can only reflect the local information between the samples,but not fully consider the shortcoming of the global information and the neighborhood information of the sample set.On the basis of the existing sparse representation,this paper combines a kind of similarity calculation method based on the shared nearest neighbor,and improves the way of the sparse subspace algorithm to construct the similarity matrix.The improved algorithm can make full use of the global and local information of the data.2.in view of the limitation that the traditional sparse subspace clustering can only express the linear relationship between samples,this thesis combines the deep automatic coder for sparse representation learning,and adds the "self-expressive layer" on this basis,making the algorithm mining the complex structure relationship between the samples and the basis of the sparse representation method.The corresponding sparse subspace clustering algorithm is proposed3.in the two application scenes of face recognition and motion sequence segmentation,this thesis selects several different high dimensional data sets to carry out experiments,and compares the results of several classical sparse subspace clustering results to show that the proposed algorithm can effectively improve the clustering results.