EEG Personal Identification Based On Recurrent Convolutional Neural Network

Brainwaves potentially have many advantages over current conventional biometrics in biometric identification field. To evaluate the feasibility of using EEG as an independent biometric modality in personal identification, this thesis studied several problems in the applications of EEG based identification based on the portable recording system. The main tasks of this thesis are as follows:1. A single-channel EEG database was collected by using portable devices,which is called the BUPT-MCPRL EEG database. The database includes EEG signals of totally 100 individuals. Five recording protocols, random device, mental state, daily routine, fatigue driving, and data augmentation were designed to explore the influences of different factors on personal identification based on EEG.2. An automatic brainwave personal identification method, ReConv-EEG Net, based on recurrent convolutional neural network (RCNN) was pro-posed. Contrary to the existing methods, ReConv-EEG Net holds four superiorities as follows. 1) The method allows testing samples of short time length, 2) The method provides a flexible mechanism to allow natu-ral and easy EEG recording protocols. 3) The method performs stable for the dataset which exists various mental states or other recording factors. 4)The method has high expansibility and potentials for multi-channel EEG signals.3. EEG motion recognition was conducted based on RCNN method. Exper-imental results on the WAY-EEG-GAL database indicate the availability of RCNN on multi-channel EEG signals.4. A real-time framework for automatic frame-level pain intensity estimation based on RCNN was constructed.