The Study Of Feature Extraction And Pattern Classification In Motor-imagery-based Brain-Computer Interface

For electroencephalography (EEG) based Brain-Computer Interface (BCI), motorimagery is considered as one of the most effective ways. It is well known thatevent-related desynchronization (ERD) can happen in the sensory motor cortex inspecific frequency bands in specific temporal segments. ERD can well be used for thediscrimination between left-and right-hand movement imagery.This paper proposes a novel temporal-frequency-spatial feature extractionalgorithm in motor-imagery-based Brain-Computer Interface, called augmented FilterBank Common Spatial Pattern. In this algorithm, electrodes, frequency bands andtemporal segments are considered to be important factors for feature extraction. Afterthe EEG measurements were bandpass-filtered into multiple frequency bands anddivided into multiple time intervals, the Common Spatial Pattern (CSP) is used for eachsignal. The two kinds of feature selection algorithms are then employed. The resultsshow that the method based accuracy yields superior classification accuracy comparedagainst the method based mutual information. The dataset from BCI competition isemployed to verify the validity of this algorithm.In addition, the present paper focuses on the selection of classifiers.Theclassification results of the linear discriminant classifier, neural network classifier andnaive Bayes classifier are compared by the extracted features. The experimental resultsshow, for the features extracted by this algorithm, the naive Bayes classifier has thebetter classification performance.This paper is cross-disciplinary study between the applied mathematics and the lifescience.