Research Of Low Rank And Sparse Representation Subspace Clustering Based On Latent Space

With the rapid development of mobile Internet,all kinds of mobile terminals will produce a lot of high-dimensional data every day,such as video image data.In order to mine the value of these data,clustering is usually used to analyze these data.However,there are a lot of irrelevant features in the high-dimensional data,so it may come across the problem of"dimension disaster"when the traditional clustering algorithm is used for directly processing these high-dimensional data.Fortunately,subspace clustering algorithm can deal with high-dimensional data very well,especially the subspace clustering algorithms based on low rank representation or sparse representation,they have attracted a lot of researchers'attention due to theirs good clustering performance.However,these algorithms also have some shortcomings,such as it is unreasonable that latent space low rank and sparse subspace clustering algorithm keeps the global structure of high-dimensional data in the latent space,and the subspace clustering algorithm is difficult to deal with non-linear structure data.In view of these shortcomings,this paper proposes some improved algorithms,the main work and innovations are as follows:?1?In order to solve the problem that it is unreasonable that latent space low rank and sparse subspace clustering algorithm keeps the global structure of high-dimensional data in the latent space,a robust latent space low rank sparse subspace clustering algorithm based on graph constraints is proposed.This algorithm can find a low rank and sparse representation and a low-dimensional latent space simultaneously,and uses graph constraints to preserve the local manifold structure of the original data to obtain discriminative latent space.In order to make the model robust to noise,this algorithm uses F-norm and7)₁ norm or7)_2,1 norm to capture noise and outlier samples.A large number of experiments show that the proposed algorithm has better performance than other subspace clustering algorithms based low rank representation or sparse representation.?2?Aiming at the problem that?1?can't deal with the data set with nonlinear structure well,a latent space deep subspace clustering model based on graph constraint is proposed by combining?1?and deep learning.In this model,F-norm and7)₁ norm are used to make the representation matrix low rank and sparse,and graph constraints are used to keep the local manifold structure of the original data.Experiments show that this model can deal with the data set with nonlinear structure well,and the latent space can keep the local manifold structure of the original data.