A method of mother wavelet function learning for DWT-based analysis using EEG signals

In brain signals analysis, there are the supplementary devices such as EEG, fNIRS, MEG, fMRI, PET, etc. EEG is a popular secondary device due to the advantages of easy usability, mobility and low-cost. Many researchers have employed a Discrete Wavelet Transform (DWT) to classify EEG signals and make a clustering of the signal in brain-computer interface and medicine diagnosis. The precision of classification and clustering for EEG analysis depend on a mother wavelet. In order to improve the precision, the previous works has taken a hand-selection method to find out the best mother wavelet after simulation. It is necessary to improve the tested precision because the best mother wavelets for the acquired EEG signals are different depending on the subjects. In this paper, we suggest a novel approach which can select the best mother wavelets for DWT-based analysis in time-series sequences of EEG signals. To show the efficiency of the proposed method, we utilized a clustering method which can separate unsupervised EEG signals into the groups such as the ADHD (Attention Deficit Hyper-activity Disorder), the normal children, and the children in the boundary between ADHD and Normal children. As a result of simulation, we confirmed that the novel method improved the precision about 15% more than the previous.