Feature Selection Method Based On Regularization Term And Sparse Representation

As an important technology of information processing and analysis,pattern recognition plays a key role in the process of mass data processing.While the ?Curse of dimensionality? exists in the progress of processing the massive data by the traditional pattern recognition technology,therefore,feature selection method has become a critical step in pattern recognition system,but also an important premise for classifier design.In this paper,we mainly study how to make full use of the information of sample data,and put forward the high performance feature selection method for pattern recognition system.The main work is as follows:(1)To exploit the spatial distribution information of similar samples in the sample data,a new feature selection method based on Lalacian regularization term called Lap-Lasso.Firstly,the redundant features are removed by using the sparse terms in the Lasso method.Secondly,a Laplacian regularization terms is introduced to preserve the spatial distribution information of the similar samples in the sample data,which can help to induce a more discriminative ability.Finally,the model of the proposed method is constructed,and APG algorithm is used to solve the optimization problem in the process.Also,the ultimate experimental results show that the proposed method has better classification performance which is compared to the traditional feature selection methods.Moreover,the proposed method has high robustness to the parameters,and it also shows better performance on different classifiers.(2)In order to exploit the geometrical information of the similar samples and the structure information of the different samples fully,a novel discriminative feature selection method called D-Lasso is proposed which is based on discriminative regularization terms.Firstly,the sparse term of the Lasso method is used to eliminate the irrelevant features.Secondly,a discriminative regularization term is introduced to keep the structure information between the geometric distribution information and the different samples.Finally,the model of the proposed method is constructed,and APG algorithm is used to solve the optimization problem in the process.And,the final experimental results show that the proposed method can promote the classification performance of the classifier,also it is robust to the parameters.Besides,the proposed method is extended to the semi-supervision.