Guo Yuxue
1.School of Health Sciences and Biomedical Engineering, Hebei University of Technology，Tianjin 300130, China； 2.Hebei Key Laboratory of Bioelectromagnetics and Neural Engineering, Hebei University of Technology，Tianjin 300130, ChinaYu Hongli
1.School of Health Sciences and Biomedical Engineering, Hebei University of Technology，Tianjin 300130, China； 2.Hebei Key Laboratory of Bioelectromagnetics and Neural Engineering, Hebei University of Technology，Tianjin 300130, ChinaYao Hang
1.School of Health Sciences and Biomedical Engineering, Hebei University of Technology，Tianjin 300130, China； 2.Hebei Key Laboratory of Bioelectromagnetics and Neural Engineering, Hebei University of Technology，Tianjin 300130, ChinaDu Bo′ai
1.School of Health Sciences and Biomedical Engineering, Hebei University of Technology，Tianjin 300130, China； 2.Hebei Key Laboratory of Bioelectromagnetics and Neural Engineering, Hebei University of Technology，Tianjin 300130, ChinaWang Chunfang
Rehabilitation Medical Department, Tianjin Union Medical Center，Tianjin 300121, China1.School of Health Sciences and Biomedical Engineering, Hebei University of Technology，Tianjin 300130, China; 2.Hebei Key Laboratory of Bioelectromagnetics and Neural Engineering, Hebei University of Technology，Tianjin 300130, China; 3.Rehabilitation Medical Department, Tianjin Union Medical Center，Tianjin 300121, China
R339.4；R496
The low classification accuracy of motor imagery EEG signals (MI-EEG) of the left and right hands limits the development of related brain-computer interface technology. The motor imagery EEG signals of 16 healthy subjects were collected experimentally. A discrete wavelet transform (DWT) and convolutional autoencoder (CAE) based classification algorithm for motor imagery EEG signals were proposed. The EEG signal is converted into a time-frequency matrix using a discrete wavelet transform and input to a convolutional autoencoder network for the feature classification of EEG signals. The algorithm obtained better classification results when tested on both the experimental dataset and the public dataset. The accuracy of the three classification groups of rest-imagine left hand, rest-imagine right hand, and imagine left hand-imagine right hand was 97.36%, 97.27%, and 86.82% on the experimental dataset, and 99.30%,98.23%, and 92.67% on the public dataset. The discrete wavelet transform combined with the convolutional autoencoder network model outperforms other deep learning methods (CNN, LSTM, STFT-CNN) in classification applications of motor imagery EEG signals of left and right hand.
