استخراج مولفه P300 ازسيگنال برانگيخته ي مغزبه منظوراستفاده درسيستم هاي واسط مغزوكامپيوتر

چكيده انگليسي :

P300 Extraction From Brain Evoked Potential For Brain Computer Interface Systems Tahere Gorji t gorji@ec iut ac ir Date of Submission 2012 03 13 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language FarsiSupervisor Rasoul Amirfattahi fattahi@cc iut ac ir Niloofar GheissariAbstractFor people who suffer from neurological diseases such as ALS who are unable to have any motor functionsbut still have some cognitive abilities Brain Computer Interfaces BCIs are a good mean to communicatewith the world through only their brain activities The main idea of a BCI is to convert the brain activities tocommands for computer or artificial limbs The main features of EEG that using in BCIs are brain rhythmsand event related potentials ERPs P300 is a component of an Event ERP which is the natural response ofthe brain after a specific stimuli with about 300 ms delay BCI development has grown fast from 1998 whenthe first BCI based on P300 so called oddball P300 Speller paradigm was introduced by Farwell andDonchin In a P300 based BCI system P300 component is usually detected by using preprocessing andfeature extraction methods to enhance signal that can be fed into a classifier to specify target or nontarget inother words with P300 or without P300 signals Since 2000 several BCI competitions have been organized in order to improve the development of BCI andunderlying data mining techniques In this thesis we propose an approach to analyze data from the P300speller paradigm of BCI competition III The objective is to predict the correct character in each of theprovided character selection epochs The user was presented with a 6 by 6 matrix of characters The user stask was to focus attention on characters in a word that was prescribed by the investigator All rows andcolumns of this matrix were successively and randomly intensified at a rate of 5 7Hz Two out of 12intensifications of rows or columns contained the desired character P300 In order to make the spellingprocedure more reliable this sequence of intensifications is repeated 15 times for each character to spell Weused two main types of features wavelet coefficients and downsampled time window and a machinelearning technique for classification Support Vector Machines SVMs SVM is a supervised machinelearning algorithm that is capable to classify linear and nonlinear separable dataset There are a number ofwavelet families with different Shapes and properties Therefore it is important to select the most suitablewavelet for analyzing the intended signal Discrete wavelet transform DWT is a mean to decompose theoriginal signal into low and high frequency components while at the same time downsampling eachdecomposed signal by a factor of 2 The high frequency wavelet coefficients are often referred to as detailand the low frequency as approximation We consider the accuracy of predicted character for differentwavelets the result shows that coif2 approximation coefficients at 4th decomposition level as a feature vectorprovides better performance than the other wavelets An important point is that the accuracy of thedownsampled method is equal to the wavelet method and equal to 97 A critical parameter in a BCI system is the time or the rate of extracting the desired signal from brain activityand convert it to commands for devices We considerd the time of predicting a character and found it takesabout 25 seconds for one character at downsampled method but wavelet method consumes only 3 seconds topredict one character That s because the feature vector size of wavelet features is 256 samples smaller thandownsampled window features This distinct difference indicates that using wavelet features and ensemble ofSVMs provided good performance Our performance is better than the winner of the BCI2005 competitions The advantage of our method is simplicity low time consuming and less computational complexity Keywords BCI EEG P300 SVM DWT

گرجي، طاهره

استخراج مولفه P300 ازسيگنال برانگيخته ي مغزبه منظوراستفاده درسيستم هاي واسط مغزوكامپيوتر

EEG , ماشين بردارپشتيبان , تبديل موجك