Single-trial EEG classification for brain-computer interface using wavelet decomposition

A classification system for EEG signals using wavelet decomposition to form the feature vectors is developed. Single-trial analysis loses the benefit of averaging to remove non-task related brain activity and makes it more difficult to pick out key features determining the execution of a task. Wavelet analysis is used here to localise the event-related desynchronization of voluntary movement. Classification of a self-paced typing experiment was made using wavelets for the feature selection and SVMs for the classification of feature vectors. Results of up to 91% classification accuracy were obtained, proving that wavelets are an effective tool, and the use of wavelets will be considered in more complex work.