Classification of hand movement imagery tasks for brain machine interface using feed-forward network

In this paper, a simple Brain Machine Interface (BMI) system that translates a change of rhythm from brain signal while performing a simulation of hand movement mentally into a real activity movement command is proposed. Four different imaginary tasks are used in the analysis process. A non-stimulus-based BCI approach is used to acquire the brain signal from ten different subjects using 19 channel EEG electrodes. Five spectral band features from each channel are extracted and associated to the respective mental tasks. The features are then classified using Feed-Forward Neural Network. The training is conducted using different ratio of training/testing data set. The developed network models are then tested for its validity. The performance of the developed network models are evaluated through simulation. The result shows that the proposed of both protocol approach and frequency sub band range selection can be an alternative general procedure to classify motor imagery task for a simple BMI system.