Eeg Feature Extraction Based On Non-negative Matrix Factorization

NMF method to achieve simple, occupy less storage space, a negative result ofdecomposition, etc. does not appear, it attracted wide attention of many scientists andresearchers. It has been widely used in various applications such as pattern recognition,information retrieval, computer vision. One of EEG data’s salient features is a largedimension and a small sample, each sample are recorded EEG signals which can bemeasured in all expression levels, but most of them has nothing to do with the samplecategories, and they don’t contain classified information, these noise will reduce theaccuracy of classification. Therefore we need to extract information about the structureand function of the brain signals from the experimental data and find each other’s contactinformation on functions, removing irrelevant information as possible. It’s the key stepthat how to extract features and reduce data dimension effectively in classification ofEEG data. In this paper, it uses the theory of non-negative matrix factorization to extractfeatures from the EEG data, and then uses the classification to verify the effectiveness andfeasibility of this feature extraction methods, the main contents are as follows:1.A feature extraction algorithm based on non-negative matrix factorization for EEGdata is proposed. The basic idea of non-negative matrix factorization is reflecting thepotential strueture of data by decomposing on non-negative matrix into the multiplicationof two. Firstly,filtered out the EEG data which needed for feature extraction.Secondly,non-negative matrix is constructed and decomposed in order to get small dimensionvectors that can fully characterize the sample.Lastly,Bayesian classifier is used tocategorize the vectors.Experimental results validate the feasibility and effectiveness of thisalgorithm.2. A feature extraction algorithm based on sparse non-negative matrix factorizationfor EEG data is proposed. Adding the sparsity constraints to the coefficient matrix iscommonly referred to sparse non-negative matrix factorization algorithm. The process is,first to filtered out the EEG data which needed for feature extraction. Secondly, sparsenon-negative matrix is constructed and decomposed in order to obtain the needed featurevectors to verify the validity of the final feature vectors extracted by classification. 3. A feature extraction algorithm based on local non-negative matrix factorization forEEG data is proposed. This algorithm is based on non-negative matrix and works byrestricting the iteration condition in three aspects:(1)Maximize the sparsity of weightmatrix H;(2) Maximize the representative of basis matrix;(3)The basis matrices shouldensure the orthogonal as possible to reduce the linear correlation and redundancy betweenthem. These three conditions emphasize the local conditions of component of the basiccharacteristics during the decomposition of the original matrix. Firstly,filtered out the EEGdata which needed for feature extraction. Secondly, local non-negative matrix isconstructed and decomposed in order to get small dimension vectors that can fullycharacterize the sample.Lastly,Bayesian classifier is used to categorize thevectors.Experimental results validate the feasibility and effectiveness of this algorithm.