An enhanced discrete wavelet packet transform for feature extraction in electroencephalogram signals

• Main Authors: Al-Qammaz, Abdullah Yousef, Yusof, Yuhanis, Ahmad, Farzana Kabir
• Format: Conference or Workshop Item
• Published: 2017
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://repo.uum.edu.my/25650/; http://doi.org/10.1145/3132300.3132303
• Institution: Universiti Utara Malaysia

Extracting features from electroencephalogram (EEG) is a challenging task because the signals are complex and chaotic in nature. EEG signals are time varying as human brain produces different frequency bands within different period of time. Due to this reason, several time-frequency methods have been used to extract features, and this includes the Discrete Wavelet Packet Transform (DWPT). DWPT was introduced to provide efficient localization of frequency bands, however, the decomposition of DWPT produces noises in the data points of sub-signals which in return affected the quality of the extracted features. Moreover, when the decomposition of DWPT happens, the length of sequences is decreased by half at every level. If it occurs at the last level, the sequence length will become very short and some frequency bands (i.e. alpha, beta, gamma) will be scattered in several location in the decomposition tree. Hence, this study introduces eDWPT which is the enhanced of DWPT. This method has been evaluated on a preprocessed EEG dataset of 6 subjects (i.e DEAP database) and compared against the standard DWPT. Two experiments of emotion recognition were performed and it is learned that the proposed feature extraction method (i.e eDWPT) produce a higher classification accuracy. Such a result indicates that the proposed eDWPT has the potential to be used as a feature of extraction method in other signal processing.