The Optimal Wavelet Basis Function Selection in Feature Extraction of Motor Imagery Electroencephalogram Based on Wavelet Packet Transformation

• Published in: IEEE access 2019, Vol.7, p.174465-174481
• Main Authors: Cheng, Liwei, Li, Duanling, Li, Xiang, Yu, Shuyue
• Format: Article
• Language: English
• Subjects: Basis functions; deep belief networks; Discrete wavelet transforms; Electroencephalography; Feature extraction; Imagery; Motor imagery electroencephalogram; root mean squared error; Root-mean-square errors; Signal processing; Signal to noise ratio; Transformations; Wavelet analysis; wavelet basis function; Wavelet domain; Wavelet packets
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2019.2953972

To solve the problem of optimal wavelet basis function selection in feature extraction of motor imagery electroencephalogram (MI-EEG) by wavelet packet transformation (WPT), based on the analysis of wavelet packet transformation and wavelet basis parameters, combine with the characteristics of MI-EEG, the characteristics of wavelet basis function suitable for feature extraction of MI-EEG are summarized. On the basis of processing and analyzing of two BCI competition data sets, signal to noise ratio (SNR), root mean squared error (RMSE), classification accuracy, and kappa value are introduced as evaluation criteria for feature extraction effect, it is concluded that the rbio2.2 wavelet basis function is the optimal wavelet basis function for feature extraction of MI-EEG. Finally, the MI-EEG collected in the laboratory is processed and analyzed, further proving that the rbio2.2 wavelet basis function is the optimal wavelet basis function for feature extraction of MI-EEG.