Motor Imagery Classification Research Based on FBCSANet Algorithm

Stroke is one of the leading causes of disability, with patients often experiencing slow and inefficient rehabilitation due to impaired motor function. With the advancement of motor imagery - based brain - computer interface (MI-BCI) technology, a novel non-invasive rehabilitation method has emerged for restoring upper- or lower-limb motor function in stroke patients. However, the low signal-to-noise ratio and non-stationary nature of electroencephalogram (EEG) signals present significant challenges for accurate classification in multi-class motor imagery tasks. In this study, we developed a deep learning model for motor imagery classification, named FBCSANet, which integrates feature extraction, multi-band decomposition, spatio-temporal convolution, and a Shuffle Attention module. The model is designed to classify three types of motor imagery tasks: left hand, right hand, and foot. EEG data from nine participants in the BCI Competition dataset were used for analysis. Experimental results show that FBCSANet achieved an average classification accuracy of 69.83 %, outperforming several baseline methods, including EEGNet and FBCSP-SVM, in both accuracy and robustness. These findings demonstrate that FBCSANet possesses reliable EEG decoding capability in multi-class motor imagery tasks, providing a theoretical foundation for the future clinical application of motor imagery - based rehabilitation techniques in stroke patients.