Motor Imagery BCI Classification Using Synchrosqueezing Transform

Developments in sensors and computer technology have made easier to record and process Electrencephalography (EEG) signals. Obtaining brain-computer interfaces (BCI) with the help of EEG signals is getting more practical and cheaper. Analyzing the meaning of the brain signals by using BCIs is popular and promising research area last years. In this work, we propose a new method to detect and classify the motor imagery (MI) EEG signals. The used dataset contains the trials for the imagination of the movements. While recording the EEG signals, subjects imagined the movements of right fist and left fist several times in each experiments. Our approach employes synchrosqueezing transform (SST) to obtain time-frequency representation matrices of EEG signals for each trial. Principal component analysis has been used for dimension reduction and feature extraction. At the end, classification process has been figured out by using support vector machine (SVM). This algorithm has been applied to each of the selected EEG channels which are directly related to motor movements. Testing and training accuracies of our SST based approach are higher than 99% for each channels separately. Also, we obtained classification results using continuous wavelet transform (CWT) to compare with the SST-based approach. Classification results of each channel have showed the promising performance of SST-based approach obviously.