A Multi-view Spatio-Temporal EEG Feature Learning for Cross-Subject Motor Imagery Classification

Abstract

This study introduces MV-FocalNet, a novel approach for classifying motor imagery from electroencephalography (EEG) signals. MV-FocalNet leverages multi-view representation learning and spatial-temporal modeling to extract diverse properties from multiple frequency bands of EEG data. By integrating information from multiple perspectives, MV-FocalNet captures both local and global features, significantly enhancing the accuracy of motor imagery task classification. Experimental results on two EEG datasets, 2a and 2b, show that MV-FocalNet accurately categorizes various motor movements, including left and right-hand activities, foot motions, and tongue actions. The proposed method outperforms existing state-of-the-art models, achieving substantial improvements in classification accuracy.