Motor Imagery Classification of Left and Right Hands Based on EEG Features and SVM

A Brain-Computer Interface (BCI) is a communication system that does not need any peripheral muscular activity. The BCI is used to translate the brain activity into a computer command. One of the most important topics in BCI is motor imagery (MI) used to show the reconstruction of the subjects. The electrical activities of the brain are measured as Electroencephalogram (EEG). EEG signals behave as low to noise ratio also shows the dynamic behaviors. In most of the existing works, the signals show the time or frequency features individually. In this work, a novel approach, based on feature extraction with discretion wavelet transform (DWT) and support vector machine (SVM) for classification is used. In this approach, both time and frequency features are extracted simultaneously which is essential to clearly understand the left hand and right hand features are used in different channels. Since the EEG signals do not have any useful frequency components above 30 Hz, the number of decomposition levels was chosen to be 4. Thus, the EEG signals were decomposed into details D5–D8 and one final approximation, A8. The features were extracted by DWT and then were classified with SVM, resulting in dramatic reduction in processing cost. By applying SVM classifier with utilization of RBF kernel, the features were categorized to separate the right hand from the left. The advantage of using DWT is captures non- stationary features. Database was recorded for BCI competition (Graz data set B), that including nine subjects, females in range of ages between 21 to 35, and males in the range of ages between 21 to 27. The sample frequency is 250 Hz. Consist of six electrodes, the first three channels are bio-polar channels and used for EEG signals and the last three ones are mono-polar used for EOG. The trail is done in five sessions, the first two sessions is data without feedback, and last three sessions is data with feedback. Each subject participates in two sessions, without feedback on two different days in two weeks. The signals are pre-processed with high-pass filtering and linear regression and classified in to two classes, including left and right hand. As using one single channel individually, may not provide us the results with desired accuracy, in the present work a combination of the channels is used. Therefore, the accuracy and specificity and permittivity for the right hand and left hand classification using SVM, increase to 71.3%, 91.89% and 64.7%, respectively. And the result illustrates that as frequency decreases, the accuracy increases to a significant amount, average accuracy of the channel Cz in beta wave approaches to 98.5%. The result shows that RBF kernel function provides significant enhancement for train and test of the SVM classification

کلیدواژه ها:

Signal Processing ، discretion Wavelet Transform ، Support Vector machine (SVM) ، Brain Computer Interface (BCI)