A Novel Motor Imagery Decoding Method Based on MODWT and Light-Weight CNN

Motor imagery (MI) is one of the most important brain computer interface (BCI) paradigms, which has been widely used in many clinical and scientific applications, such as rehabilitation and motion control. Despite the fast development of deep learning method, MI electroencephalography (EEG) signal decoding has remained a critical challenge to because of its poor SNR, low spatial resolution, large inherent variability, and small dataset size. In this work, we propose a maximal overlap discrete wavelet transform (MODWT) and light-weight convolution neural network (CNN) combined method, which effectively improves EEG decoding performance with relatively fewer network parameters by incorporating time-frequence and time-spatial features of EEG. Besides, the method employs channel attention mechanism to enhance important features and reduce redundance. We conduct the experiment on the public dataset BCI Competition IV 2a, and the results show that the proposed method can achieve better performance than other light-weight model, with 78.7% accuracy (Kappa 0.71) with four-class classification, which well validates the effectiveness and robust of the proposed method.