Local-Global Multi-Temporal-Spatial Attention Network for Motor Imagery Classification

This paper proposes a novel network designed to tackle the challenges associated with low spatiotemporal resolution and the lengthy sample times required for effective decoding in motor imagery (MI) Electroencephalogram (EEG) classification. The proposed approach employs a Local-Global Multi-Temporal-Spatial Attention Network (LGMA-Net), which leverages advanced attention mechanisms to simultaneously capture both local and global features within the spatiotemporal domain. By integrating multiple spatiotemporal attention strategies, the network effectively processes motor imagery EEG signals, enhancing the extraction of spatiotemporal features. Results from the BCI IIa dataset demonstrate strong real-time decoding performance, requiring shorter input signal lengths to achieve improved decoding accuracy. The results from the BCI IIa dataset demonstrate strong real-time decoding performance, utilizing shorter input signal lengths to achieve improved decoding accuracy.