Application of Genetic Algorithm and Wavelet Transform in Eye-Blink Detection from Electroencephalograms
Authors: Roy Francis Navea and Elmer Dadios
Abstract
This paper presents an algorithm for detecting eye-blink artefacts in an electroencephalogram (EEG) signal using optimized wavelets. The algorithm is based on genetic algorithm optimization techniques and wavelet transform. A mother wavelet was made to fit the sought after waveform in the EEG signal using genetic algorithm (GA). A mother wavelet, the Shannon (Sinc) wavelet, was deliberately altered using a fitness function described by significant contributing parameters used for translating and scaling the wavelet. Once the optimized wavelet is obtained, it was made to run through an EEG signal to search for likeliness which is determined by a threshold correlation value obtained using shape language modeling (SLM). Other methods were considered in detecting the eye-blinks such as the manual observation and the kurtosis method. Results show a significant difference in the counting results of the three methods covered by this paper. As for the proximity of counts, the wavelet method counts eye blink signals closer to the manually observed counting. Results also show a significant difference between the wavelet-based and the kurtosis-based eye blink counting methods.