Research on EEG Signal Classification Method Based on Brain-Computer Interface Control of Rehabilitation Robot

Brain-computer interface (BCI) control of multiple rehabilitation robots provides a novel type of human-robot interaction and an important research direction in the field of intelligent rehabilitation. However, most current EEG signal classification algorithms for controlling rehabilitation robots exhibit several drawbacks, including a lack of novelty, insufficient classification accuracy and poor adaptability to different experimental paradigms. Therefore, we designed a new deep-learning method to classify EEG signals to enhance the effectiveness of BCI control of rehabilitation robots. Here we apply this method to classify EEG signals and control the rehabilitation robot to accomplish a target task. In addition, we compare our algorithm with several different EEG signal classification algorithms in two experimental paradigms: motor imagery and motor execution. Our deep learning method has the highest offline accuracy of 72.7% and an average accuracy of 72.4% under the motor imagery experimental paradigm, and the highest offline accuracy of 97.5% and an average accuracy of 90.4% under the motor execution experimental paradigm. These results show that our EEG signal classification algorithms is more effective compared to other EEG signal classification algorithms and can effectively be used to control a rehabilitation robot across a wider range of applications.