 This paper proposes a novel multi-scale hybrid convolutional neural network, MSH-CNN, for decoding motor imagery electroencephalogram, e.g. signals to improve classification performance. The two-dimensional convolution is used to extract temporal and spatial features of e.g. signals, while the one-dimensional convolution is used to extract advanced temporal features of e.g. signals. Furthermore, a channel coding method is proposed to enhance the expressive capability of the spatial temporal characteristics of e.g. signals. The proposed method was evaluated on the dataset collected in the laboratory and BCI competition 4-2B, 2A, and the average accuracy was found to be 96.87%, 85.25%, and 84.86%, respectively. Compared with other advanced methods, the proposed method achieved higher classification accuracy. Moreover, the proposed method was applied to an online experiment and a smart prosthetic arm control system was designed. The results showed that the proposed method successfully extracted e.g. signals advanced temporal and spatial features, and the online recognition system contributed to the further development of the BCI system. This article was authored by Xianluan Tang, Kaikuan Yang, Xia Sun, and others. We are article.tv, links in the description below.