Convolutional LSTM: A Deep Learning Method for Motion Intention Recognition Based on Spatiotemporal EEG Data

Abstract

Brain-Computer Interface (BCI) is a powerful technology that allows human beings to communicate with computers or to control devices. Owing to their convenient collection, non-invasive Electroencephalography (EEG) signals play an important role in BCI systems. Design of high-performance motion intention recognition algorithm based on EEG data under cross-subject and multi-category circumstances is a crucial challenge. Towards this purpose, a convolutional recurrent neural network is proposed. The raw EEG streaming is transformed into image sequence according to its location of the primary sensorimotor area to preserve its spatiotemporal features. A Convolutional Long Short-Term Memory (ConvLSTM) network is used to encode spatiotemporal information and generate a better representation from the obtained image sequence. The spatial features are then extracted from the output of ConvLSTM network by convolutional layer. The convolutional layer along with ConvLSTM network is capable of capturing the spatiotemporal features which enables the recognition of motion intention from the raw EEG signals. Experiments are carried out on the PhysioNet EEG motor imagery dataset to test the performance of the proposed method. It is shown that the proposed method can achieve high accuracy of 95.15%, which outperforms previous methods. Meanwhile, the proposed method can be used to design high-performance BCI systems, such as mind-controlled exoskeletons, prosthetic hands and rehabilitation robotics.

This work is supported in part by the National Key R&D Program of China (Grant 2018YFC2001700), the Strategic Priority Research Program of Chinese Academy of Science (Grant No. XDB32000000), and Beijing Natural Science Foundation (Grant L172050 and 3171001).