Comparative Study Between Daubechies and Coiflets Wavelet Decomposition Mother Families in Feature Extraction of BCI Based on Multiclass Motor Imagery Discrimination

Abstract

Brain-computer interface (BCI) is a method of direct communication between the human brain and the computer without the need for the use of other physical organs such as peripheral muscles, where the interface allows the human to order any electronic device using brain activity only. Multiclass discrimination of motor imagery movements is one of the BCI applications that uses electrical signals measured from brain or muscles; these signals were filtered and processed using time, space, and frequency feature extraction techniques, and machine learning algorithms to construct a system achieve best performance. The objective of this research was to make a comparative study between different wavelet Daubechies and Coiflets families mother functions used as feature extraction to discriminate between 4 classes of imagined movements. The coefficients of different Wavelet Daubechies (2-20) and Coiflets (1-5) were examined in calculating common spatial filters to discriminate between 4 imagined classes, where 4 detail levels of the wavelet coefficients were calculated; support vector machine was used as a classifier that enables the classification of imagined movements. The algorithms applied to the BCI competition IV data set 2a. Result yields performance of 0.53 and 0.52 average κ coefficients using Db2 and Coif1, respectively.

Discrimination of Multiclass Motor Imagery-Based Brain-Computer Interface

“…The first 3 results of this technique is much better than that of Elmahdi et al's technique 8 because time windowing is considered to be important computing CSP; moreover, the last 3 subjects were slightly different, possibly because an inefficient rejection technique was used.…”

“…Table 1 shows the results obtained by the proposed algorithm and the results obtained by Elmahdi et al's algorithm. 8 By comparing the results of this system, it is obvious that the optimum average value that can be achieved is 0.638, as shown in Table 1. This performance has been done using wavelet coefficients to calculate CSP features in different time segments.…”

“…The proposed algorithm applied on 6 subjects from the Data Set II A from BCI competition IV. Table 1 shows the results obtained by the proposed algorithm and the results obtained by Elmahdi et al's algorithm 8 …”

“…Elmahdi et al's 8 research indicated that the suitable way for feature extraction from the (fifth order Chebyshev type II, 8–30 Hz) IIR band pass filtered data is using the time-frequency domain methods such as wavelet transform (WT), as they proposed that previous literature has proven that discrete wavelet Daubechies (db) and Coiflet (coif) families are the most suitable families for similar applications which are used to decompose the recorded data (S) into multiresolution subsets of coefficients; a detailed coefficient subset (cDi) and approximation subset (cAi) at the level. Researches showed that the differences of the frequency values in mu (μ) and beta ( β ) rhythms are more apparent according to event-related desynchronization/event-related synchronization features, which are associated with brain motor output.…”

Modeling and Causality Analysis of Human Sensorimotor Control System Based on NVAR Method

“…0≤α≤1 α = 0.5 where is a penalty parameter and is set as [22]. By setting this sparsity-inducing term, the data dimension of the EEG-EMG signal and the complexity of NVAR model are reduced [23].…”