Motor Imagery BCI System Using Bidirectional Long-Short Term Memory and Discrete Wavelet Transform

• Main Authors: SHE,PING-HUNG, 佘秉宏
• Other Authors: LIN,JZAU-SHENG
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/sw5a3f

碩士 === 國立勤益科技大學 === 資訊工程系 === 107 === Motor Imagery is a method to generate EEG signals by using imaging proper activities such as moving left- or right- hand, legs and so on. These EEG signals can be used to control peripheral devices through a Brain Computer Interface (BCI) system. How to classify signals after imaging is an important issue. In order to make full use of the time-frequency characteristics of sequence signals in EEG, a deep learning model modified from Long Short-Term Memory (LSTM) is applied. The Bidirectional LSTM is used to classify the signals. This article recorded four kinds of brainwave signals, which are imagined left-handed, imagined right-handed, imagined tongue and blinking. This article combined the g.SAHARAsys system with Emotiv EPOC IC to extract brainwave signals and then wirelessly transmit them to a computer. According to the international 10-20 system, we use three electrode points, C3, C4 and Cz, to intercept the brain wave signal, and then use the Discrete Wavelet Transform (DWT) to extract the time-frequency characteristics of the motional imaginary brain wave. Finally, the motion imaginary signals were identified through the Bidirectional LSTM.