A self produced mother wavelet feature extraction method for motor imagery brain-computer interface

Motor imagery base brain-computer interface (BCI) is an appropriate solution for stroke patient to rehabilitate and communicate with external world. For such applications speculating whether the subjects are doing motor imagery is our primary mission. So the problem turns into how to precisely classify the two tasks, motor imagery and idle state, by using the subjects' electroencephalographic (EEG) signals. Feature extraction is a factor that significantly affects the classification result. Based on the concept of Continuous Wavelet Transform, we proposed a wavelet-liked feature extraction method for motor imagery discrimination. And to compensate the problem that the feature varies between subjects, we use the subjects' own EEG signals as the mother wavelet. After determining the feature vector, we choose Bayes linear discriminant analysis (LDA) as our classifier. The BCI competition III dataset IVa is used to evaluate the classification performance. Comparing with variance and fast Fourier transform (FFT) methods in feature extraction, 2.02% and 16.96% improvement in classification accuracy are obtained in this work respectively.