Multi-class imagery EEG recognition based on adaptive subject-based feature extraction and SVM-BP classifier

In brain-computer interface (BCI), the classification accuracy significantly drops when the system has multiple motor imagery tasks for different subjects. To improve the classification accuracy and individual adaptability of the system, a new method of feature extraction and classification is presented in this paper for recognizing the four different motor imagery tasks (right hand, left hand, foots and tongue movements). Wavelet packet basis is selected for extracting the frequency bands in which the feature can be classified easily. Then, the EEG feature is extracted from the frequency bands information with One Versus the Rest Common Spatial Patterns (OVR-CSP) algorithm. Furthermore, a hybrid classification model of Support Vector Machines combining with the Back Propagation neural network (SVM-BP) is built to classify the multi-class EEG feature. Experiment results show that the proposed approach achieves better performance than other methods and it has adaptability for different subjects to some extent.