Multivariate Iterative Filtering-Based SSVEP Detection in Mobile Environment for Brain–Computer Interface Application

This letter aims to improve the steady-state visual evoked potential (SSVEP) detection performance in mobile environments. Multiscale analysis of the eight-channel electroencephalogram (EEG) signals is performed using multivariate iterative filtering (MIF). Mode-aligned multivariate modes obtained from MIF are fed to canonical correlation analysis (CCA) for finding the correlation with sine–cosine reference signal. The correlation coefficients from different multivariate intrinsic mode functions are computed as features, which have been classified using machine learning classifiers: k nearest neighbor, linear discriminant analysis (LDA), and support vector machine (SVM). The proposed framework is evaluated using a real-time EEG dataset recorded in a mobile environment with the help of extensive experiments. The LDA classifier provides 88.99%, 84.13%, 81.52%, and 76.62% accuracies for 0.0, 0.8, 1.6, and 2.0 m/s speed, respectively, when classifiers are trained specific to each subject. Subject-independent LDA classifiers achieve 89.49%, 85.00%, 84.20%, and 69.90% accuracies for the aforementioned four different speeds. The MIF-based CCA (MIF-CCA) framework achieved slightly higher accuracy than conventional CCA-based SSVEP detection when the subject was standing or moving at a lower speed, but when the subject was moving at a speed of 2.0 m/s, the average accuracy of MIF-CCA was higher by 21.86%, as compared with CCA algorithm, which shows the usefulness and robustness of the proposed approach. Finally, the proposed feature extraction techniques for mobile EEG signals will be useful for classifying EEG signals in a mobile environment.