An Automatic System for Characterization and Detection of Ocular Noise

Abstract

Eye blinks cause high amplitude noise in electroencephalograms (EEGs), the noise from these blinks causes interference in several very important frequency bands. The method detailed in this paper uses independent component analysis and a diversified feature space to identify and filter out eye blink noise during wearable electroencephalographic tests. Prior work used autoregressive modeling in the time domain to identify blink segments in the recorded data. While the previous autoregressive method showed high accuracy in short trials, the goal of this work is to create a more advanced system capable of filtering blink noise in long, continuous trials. One of the major applications for this system is improving the quality of data collected during workload assessment tasks. Trials that consider the subject’s workload over time involve sensitive calculations done over the long term, and blinking resides in frequency bands that are known to be useful in determining the subject’s csurrent workload. A blink in one of these bands could give a false positive result for workload, or it could confuse an algorithm during training. In smaller studies subjects have been told not to blink, or were told to keep their eyes closed, but for workload assessment tasks it’s usually not practical to tell the subject to not blink during a strenuous trial. Other methods have been introduced that involve electrooculogram (EOG) data; the proposed system only uses electrooculogram data for training purposes, after this channels can be removed, so that wearable system scan reduce the amount of data recorded per second.