An improved MI recognition by localising feature extraction in both frequency and time domains

ABSTRACT

In EEG-based Motor Imagery (MI) recognition, traditional features are usually extracted in time-frequency domain without localisation. Our experimental results indicate that localisation in either frequency or time domain can alone improve the MI recognition task. In this study, we have developed a frequency-time localised feature extraction (FTLFE) technique that boosts the performance further when compared to localisation in any single domain. This finding is verified by using two standard datasets. We have also shown that the proposed FTLFE method is robust against different parameters. A comparison with a number of traditional features and methods is presented that corroborates the superiority of the proposed method.

KEYWORDS:

• EEG-based Motor Imagery (MI)
• Brain Computer Interface (BCI)
• frequency-time localised feature extraction

Acknowledgements

This study was sponsored by the United International University (reference: UIU/IAR/02/2019-20/SE/11).

Disclosure statement

No potential conflict of interest was reported by the authors.