Accuracy electroencephalography classification by a regularized long short-term memory network

Abstract

The motor-based brain–computer interface is widely used in the exoskeleton reconstruction of patients with muscle weakness and to enhance the operating experience of somatosensory game customers through the combination of actions and electroencephalography signals. However, the recognition algorithms in traditional motor-based brain–computer interfaces have problems such as “brain–computer interface blindness” (recognition accuracy is less than 70%) and “one person one model.” In this study, a regularized long short-term memory algorithm and a hardware platform for gesture recognition by using the motor-based brain–computer interface are proposed. Experimental results show that the gesture recognition accuracy rate based on the motor brain–computer interface is up to 95.69%, which is significantly better than that of other algorithms. The proposed model enhances the applicability and generalization ability of the brain–computer interface, for which the practicability and effectiveness are verified.