Font Size: a A A

Research On EEG Feature Extraction Based On Multiple Linear Algebra

Posted on:2015-04-16Degree:MasterType:Thesis
Country:ChinaCandidate:L YangFull Text:PDF
GTID:2298330422470753Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
Brain-computer interface (BCI) based on EEG is a direct pathway between biologicalbrain and an external device, organism can transfer human information and controlexternal equipment directly through EEG signals.It has great significance and applicationprospects in medicine and other fields. EEG feature extraction is an important steptowards pattern recognition.This paper has improved the traditional feature extraction algorithm for theapplication background of EEG signal processing and do the following tasks based onmulti-linear algebra:First, this paper has used a multi-linear principal component analysis (MPCA)framework based on tensor feature extraction for the limitations of traditional principalcomponent analysis (PCA) and two-dimensional principal component analysis (2DPCA)when they handling multi-channel EEG. It reduced dimensionality and extracted featuresof EEG data by the projection of tensor to the matrix. While the second and third EEGdata tensor input is used in simulation experiments; this method has solved the bottleneckproblem of high dimensionality, and improved the recognition accuracy effectively.Secondly, this paper has used non-correlated multi-linear principal componentanalysis (UMPCA) framework for the traditional PCA,2DPCA of processing EEG data inmulti-level input problems and improved initialization original method. It diddimensionality reduction and extract non-relevant characteristics for the second andthird-order input EEG data by using tensor to vector projection. This method takes intoaccount the correlation between the features and more closely with respect to therecognition process MPCA method, while addressing the EEG data processing problemsof small samples.Finally, taking into account the limitations of multistage input in the traditionaldiscriminant analysis (LDA) for EEG recognition problem, this chapter outlines themulti-linear discriminant analysis (MLDA). Then we have improved the non-correlatedmulti-directional discriminant analysis method (UMLDA) on this basis and addedregularization parameters r. At last the third and fourth order EEG data input were donedimensionality reduction and feature extraction of non-related. The experimental resultshave fully reflected the practicality of non-relevant features, while adding regularizationparameter adapt to each group input and improving the recognition accuracy; also it has solved the curse of dimensionality and the small sample size problem effectively.
Keywords/Search Tags:EEG Signals, Feature Extraction, Tensor, Principal Component Analysis, Multi-linear Principal Component Analysis, Linear Discriminant Analysis, Multi-linear Discriminant Analysis, Non-correlated Multi-linearDiscriminant Analysis
PDF Full Text Request
Related items