 This paper proposes a novel approach to accurately decode motor imagery, MI, base-brain-computer interfaces, BCIs. It combines a multi-branch spectral temporal convolutional neural network, MBST-CNN, with a light-GBM model to extract features from EEG signals. This method outperforms existing approaches in terms of accuracy, demonstrating its potential for use in MI-based BCI applications. This article was authored by Hai Jia, Shi Chi Yu, Xuan Jia Yin, and others.