Spectral Power Estimation for Unevenly Spaced Motor Imagery Data

The human brain can send a command to external devices or communicate with the outside environment by the means of a brain computer interface (BCI) system. The effectiveness depends on how precisely specific brain activities can be identified from EEG. Noise is usually mixed into the EEG signal, and cannot be separated or filtered out in some cases. In a practical BCI system, the whole segment of EEG is discarded when a portion of that segment is contaminated by extreme noise or artifacts. This leads to the weakness that the BCI system cannot output decoding results during the period of that discarded segment. In order to solve this problem, we employed a Lomb-Scargle periodogram to estimate the spectral power based on an unevenly spaced segment, a portion of which has been removed due to noise contamination. According to the classification results of motor imagery data, the accuracy is not dramatically decreased along with increased proportion of data removal.