A fusion wavelet-based binary pattern approach for enhanced electroencephalogram signal classification

Motor Imagery- Brain-Computer Interface (MI-BCI) enables people with disabilities to communicate through a channel that does not require them to use their muscles. We present a comprehensive and robust feature extraction technique for Electroencephalogram (EEG) data to reach the optimal degree of classification accuracy through fusion methodologies. Three principal techniques are employed for processing the EEG data: the Wavelet Packet Decomposition (WPD) methodology, the Discrete Wavelet Transform (DWT) technique, and the Quad Binary Pattern (QBP) method. The proposed approach comprises five distinct methods, including two fusion operations that leverage both EEG signals and pattern-based techniques. To extract histogram features, one hybrid approach involves decomposing EEG signals using DWT followed by transforming resultant data into the QBP domain. Another model employs the decomposition of signals using WPD, with the output subsequently transferred to the QBP domain for feature extraction. For classification, our study makes use of an Artificial Neural Network (ANN). The proposed framework was validated by the binary class, which is comprised of the imagined movement of both the left and right fists. A DWT-based QBP approach attains a degree of accuracy on the PhysioNet dataset that was 99.50 %.