ELSTM-Based Visual Decoding from Singal-Trial EEG Recording

摘要

Electroencephalograph (EEG)records spontaneous electrical activity in the human brain, which can be utilized to read the human mind. The study works to decode the content of object image from EEG recorded while the subjects are looking at images. The identification system is developed based on the Long Short-Term Memory (LSTM). Inspired by the cognitive science, features from multiple stages in LSTM network are utilized to discriminate EEG. The proposed method reduces the dependence on the high-density recording EEG and obtains an accuracy of 96.2% for 40 categories. The good performance indicates that the proposed system can be applied to Brain-Computer-interface (BCI) in the future.