A Novel Classification Model Based on DWT and CNN-LSTM Motor EEG Imagination Signals

Feature extraction and pattern classification play important roles in brain-computer interface (BCI) systems, which are widely used in rehabilitation medicine, artificial intelligence and other fields. Although advanced EEG acquisition technology can generate a large amount of EEG data for different brain regions, it inevitably has disadvantages such as high cost, long time consuming and inherent high false positives. Considering the contradiction between data volume and accuracy, this paper proposes a MI EEG feature extraction and classification method based on discrete wavelet transform (DWT) and Convolutional Neural Network-Long Short Term Memory (CNN-LSTM). Based on the project BCI motor imagination EEG(MI-EGG) data set, all evaluation indexes of the proposed CNN-LSTM model are superior to other models. The prediction accuracy, precision, recall rate, F1-score and MCC of C3 and C4 channels reaches 96.43%, 96.57%, 96.57%, 96.57% and 93.02%, respectively. The reliability of the proposed model is also proved by tenfold cross-validation. The average accuracy is 99.64%, which is far higher than the other models. It can be concluded that it is a reliable method to predict MI-EEG signals by using CNN-LSTM to train the modeling ability of images.