[Feature extraction and classification of EEG for mental tasks based on wavelet packet analysis]

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Electroencephalogram (EEG) signals are crucial for understanding brain activity.
• Analyzing spontaneous EEG for mental task identification presents challenges in feature extraction.
• Existing methods like autoregressive models have limitations in capturing complex EEG dynamics.

Purpose of the Study:

• To investigate the efficacy of wavelet packet analysis for feature extraction from spontaneous EEG.
• To compare the performance of wavelet packet analysis against autoregressive models for mental task classification.
• To assess the potential of wavelet packet analysis as a robust method for EEG signal processing.

Main Methods:

• Artifact-free EEG segments were processed using dyadic wavelet packet decomposition.
• Feature vectors were generated from the energy values of different EEG component subspaces.
• Radial basis function networks were employed to classify three distinct mental task pairs.

Main Results:

• Wavelet packet analysis demonstrated significantly higher classification accuracies than the autoregressive model method.
• The multi-scale representations derived from wavelet packet decomposition proved effective for distinguishing mental tasks.
• Feature vectors based on energy values captured relevant EEG signal characteristics.

Conclusions:

• Wavelet packet analysis is a promising technique for extracting discriminative features from spontaneous EEG signals.
• This method offers improved performance over traditional autoregressive approaches for mental task classification.
• The findings support the adoption of wavelet packet analysis in brain-computer interfaces and neurological studies.