Research And Implementation Of Motor Imagery EEG Analysis Algorithm

Brain-computer interface,refers to a communication system which is built between human brain and computer or other electronic equipment directly.By collecting and measuring EEG signals that was produced from human brain,BCI outputs the corresponding command through feature extraction and classification recognition.This way of communication could offer a help to people whose brain function is totally good but cannot effectively control the muscle activity because of trauma and nerve disease.Brain-computer interface plays a very important role in medical,rehabilitation training,military training,industry,and has become a research hot spot in recent years.This thesis mainly aims at the intelligent need of current BCI system,and provides several ways of feature extraction and classification recognition.The thesis primarily finished the following work:In the stage of EEG signals' processing,through the spectrum analysis to the EEG signals that were collected under subject's four kinds of imagination and resting state of brain,the optimal filtering spectrum could be properly confirmed combined with the ERD/ERS of every single motor imagery.In the research work of feature extraction method,two sets of feature extraction scheme was implemented in view of the four types of motor imagery.The first set of scheme extracts the most obvious spectrum of EEG features through DWT,then reduces the dimension of feature sets by using PCA thus improves the efficiency of classification.The second set of scheme designs two different ways of feature extraction including OVR-CSP and PW-CSP which is based on traditional CSP algorithm solving two types of problems.According to the principles of SVM dealing with the four-class classification problem,combined the two sets of feature extraction scheme with SVM to implement the classification of four types EEG.Then simulate the whole process of feature extraction and feature classification in Matlab.The simulation results show that the highest classification accuracy of first set of feature extraction scheme reached 82.3%,the highest accuracy of the second set of feature extraction scheme is 68.7%and 78.1%corresponding to OVR-CSP and PW-CSP.So we could draw the conclusion that PW-CSP is more effective than OVR-CSP in feature extraction for this data set and the two sets of feature extraction scheme is feasible due to the classification accuracy.After implementation of the whole feature extraction and feature classification in Matlab,finally,the thesis also discusses the design of related feature extraction algorithm,including the module design of Mallat algorithm and principal component analysis algorithm in Quartus II,the simulation of these modules is conducted as well.