Multi-channel EEG signals classification via CNN and multi-head self-attention on evidence theory

Electroencephalography (EEG) provides valuable physiological information to identify human activities. However, it can be difficult to analyze the EEG data in human patterns identification, because both subjective and objective factors can easily affect sensitivity. In this study, a novel multi-head self-attention convolutional neural networks (CNN) framework based on Dempster-Shafer (D-S) evidence theory, called ETNN, is proposed to classify the EEG signal. The ETNN model considers the multi-type networks and fuses multi-output with D-S evidence theory, which can handle the EEG data more reasonably. In particular, a classification algorithm for EEG signals is derived with information fusion. Finally, an application for event-related potential signal classification and sensitivity analysis is used to demonstrate the effectiveness of the proposed ETNN model compared with existing classification techniques.