Pattern recognition of motor imagery EEG signal in noninvasive brain-computer interface

Abstract

Brain-Computer Interface (BCI) enables people to communicate with the external objects directly only by using electroencephalography (EEG) rather than other central nerves, peripheral nerves or effectors in brain normal pathway. Spontaneous and noninvasive, Brain-Computer Interface based on motor imagery EEG signal can be applied more widely. This paper focuses on pattern recognition of EEG signal in Brain-Computer Interface based on the left-right hand motor imagery. Mental task among 5 subjects was processed to acquire motor imagery EEG signal. In order to improve the signal-to-noise ratio (SNR) of EEG, filters in time domain, frequency domain and special domain are applied in signal preprocess. Additionally, power spectral density (PSD) was extracted with sliding windows from processed signal as features for pattern recognition. Then the discriminate labels of motor imagery are trained by linear discriminate analysis (LDA) and random forest algorithm to decode the imagine pattern. The results show an average recognition accuracy of 0.65 ± 0.07 for LDA and 0.70 ± 0.05 for random forest can be obtained, which has significant difference with random level. From the results, it can be concluded that pattern of motor imagery can be decoded. In addition, a signal trial online pattern recognition task can be further implement to achieve human-computer interaction based on EEG signals.