Wavelet based feature extraction for classification of motor imagery signals

The analysis of EEG signals play a significant role in brain related studies. Accurate investigation and analysis of the EEG signals from a subject can interpret the inherent information about the intention of the person to some extent. The accuracy of such interpretation or detection can be of utmost importance for various brain computer interface (BCI) based applications. For instance, within the last decade, BCI has been widely investigated and employed to assist in the restoration of sensory and motor functions in paralyzed individuals. EEG typically contains numerous information about the cognitive thinking and intention of a person, particularly in terms of motor movements. Due to this, comprehensive methods of EEG signal arrangement and pattern recognition through signal classification techniques are required to precisely predict the intended set of movements. In this study, an EEG classification technique based on a combination of wavelet transform analysis and Neural Networks (NN) is presented. Daubechies wavelet decomposition (db8) has been employed to decompose the recorded EEG signals into four levels. These decomposed level details are then used to calculate the feature set which is input to NN classifier for further classification. A total of 6 features were used to perform feature wise classification, where Integrated EEG (IEEG) feature set has been found to possess the highest classification accuracy of 89.39 % with a NN classifier of 9 hidden layers. Whereas, a classification accuracy of 94.86 % was achieved when the features were arranged and cascaded horizontally in form of a dataset as input to a NN classifier of 5 hidden layers.