The effects of pre-filtering and individualizing components for electroencephalography neural network classification

Abstract

Brain Computer Interfaces use electrical impulses from the brain to control hardware to perform a task. Traditionally these electrical impulses are recorded in the form of electroencephalography with electrodes placed on the subject's scalp. This paper is focused on processing the electroencephalography signal using independent component analysis to train a neural network to recognize left and right hand grasping motions. The independent component analysis was used to reduce noise and isolate the recording channels that are associated with muscle movement. The performance of the network is compared by training the neural network with and without the pre-preocessing independent component analysis and evaluating the rate at which the system identifies the correct grasping motion through the baseline noise present in electroencephalography recordings.