NeuroMotion: A lightweight and Portable Rehabilitation Training System for Motor Imagery

In response to the problems of poor portability and low decoding efficiency of the current rehabilitation system based on motor imagery(MI). In this paper, we propose a lightweight and portable MI-based rehabilitation training system named NeuroMotion, which consists of two parts: signal acquisition and signal decoding. Signal Acquisition: Only 8 channels are used to acquire MI EEG signals to improve the portability of the system. Signal decoding: A Triple-Branch Shallow Convolutional Neural Network (TBSCNN) model is constructed to achieve efficient decoding of EEG signals from motor imagery; branches 1 and 2 of the TBSCNN consist of one layer of temporal convolution and one layer of spatial convolution, while branch 3 consists of one layer of temporal-space fusion convolution. The three branches extract EEG spatio-temporal features in parallel and finally output the categories after one universal convolutional layer. Experimental results show that the NeuroMotion proposed in this paper improves the portability and decoding efficiency of the rehabilitation system. The signal decoding model TBSCNN achieves 84.37% MI EEG binary classification accuracy on the self-acquisition dataset. The lightweight EEG decoding model and the setup of a small number of EEG acquisition channels lay the foundation for the development of a low-power, portable brain-computer interface rehabilitation device.