Multi-mode EEG Pattern Recognition And Application Systems

Human brain is one of the most complex components during the process of evolution in nature.Human brain provides human language,memory,cognition,emotion and other advanced information processing functions for human beings.Since the 21 st century,one of the most significant challenges is how to study and reveal the functional mechanism of brains.Brain-computer interface(BCI)establishes a communication between brain and computers or other electronic devices by extracting specific brain activity instead of using traditional peripheral nervous system.As a typical multidisciplinary research field,BCI is one of the most rapidly developing research field.The fundamental problem of brain-computer interaction is how to extract robust features from time serials data,and to classify the patterns through machine learning and pattern recognition methods.However,due to low signal-to-noise ratio of EEG signals,existing BCI technology can’t provide a stable system with high-performance.This paper aims to study novel methods of EEG data preprocessing,feature extraction,and classification based on the spatial-frequency-spectrum characteristics of EEG signals.Finally,we develop a novel system based on BCI technology,and focus on studying the brain activity pattern of specific populations.The main contributions of this paper are given as follows:1.High-order canonical correlation analysis method for EEG classification: A tensorbased method,called Higher-order Canonical Correlation Analysis(HOCCA),is proposed with the aim of simultaneously seeking individual spatial and spectral subspaces for each class so that each class is projected into its own subspace separately.The experimental results demonstrated its superior performance.2.Uncorrelated tensor-based nearest feature line method(UTNFL)for EEG classification: This paper proposes a novel uncorrelated tensor-based nearest feature line method for feature extraction and classification.UTNFL takes high-dimensional spatial-frequency-spectrum characteristics and the correlation between features into consideration,and extracts low-redundancy EEG features.Extensive experiment comparisons have been performed on three datasets.The experimental results demonstrate the superior performance of the proposed method over the contemporary methods.3.Common Spatial-Spectral Boosting Pattern(CSSBP): Among various approaches developed for EEG signals,common spatial patterns(CSP)has been proved to be one of the most effective algorithms.However,studies have shown that the performance of CSP algorithm heavily depends on its operational frequency bands and channel configuration.However,there is no agreed upon clinical conclusion about motor imagery patterns of stroke patients.In this paper,we propose an adaptive boosting algorithm,termed common spatial-spectral boosting pattern(CSSBP),to promote the performance of decoding EEG patterns from stroke patients by a simultaneous optimization of the frequency filter and spatial filter.The most important channel groups and frequency bands related to motor imagery can be extracted by our algorithm.4.BCI-FES motor function rehabilitation system for stroke patients: In this study,we design and develop a novel Brain Computer Interface-Functional Electrical Stimulation(BCI-FES)based rehabilitation training platform for stroke patients.We collect some real stroke patients and divide them into the experimental group and control group.Results show that our system can effectively promote the rehabilitation of stroke patients.5.Exploration and analysis of motor imagination patterns of stroke patients: The desynchronization potentials evoked by motor imagery of patients with brain lesions are quite different from the ones by motor imagery of normal persons.The differences attribute to spatial locations,frequency and amplitude.In this paper,we focus on developing active induced motor imagery paradigms and developing algorithms for recognizing the desynchronization potentials evoked by motor imagery of patients with brain lesions.Furthermore,we aim to reveal the neurophysiologic mechanism of brain motor functional reconstruction during rehabilitation.In summary,this paper explores the motor imagery EEG patterns of both normal people and stroke patients.We propose a number of feature extraction and classification methods for EEG signals in spatial-spectral-temporal modes,improving the performance of BCI technology and its applications in real scenarios;We design and develop a BCI-FES rehabilitation system for stroke patients,which promotes patients to participate in rehabilitation training actively,and thus improves rehabilitation performance for stroke patients;We utilize machine learning and pattern recognition methods to explore the rehabilitation mechanism of stroke patients during motor imagination,providing theoretical knowledge for stroke rehabilitation.