부산대학교 도서관

Brain-computer interface (BCI) is a tool that enables direct communication with a computer using neural activity as a control signal. The most widely known electroencephalograms (EEGs) include steady state visual evoked potential (SSVEP), P300 [1], and motor imagery [2]. Among them, SSVEP extracts features with relative ease and has high classification accuracy; however, it requires a visual stimulus device, such as a liquid crystal display (LCD) or light emitting diode (LED). In practice, numerous studies have used LCDs as the visual stimulus device. However, using an LCD for actual use blocks the subject's forward view, and it lacks portability, thus preventing the construction of a wearable BCI. In addition, the following conditions must be met to construct a BCI that is less burdensome for the users. (1) The delay in system response must be very short; (2) the BCI must be highly portable; (3) it must have low power consumption. This study attempts to overcoming the issues of blockage of forward vision and portability by presenting visual stimuli using smart glasses. As smart glasses can display images overlaid on the actual view, the field of vision is not blocked, and as they are lightweight, they are highly portable and wearable. Previously, some studies have used smart glasses as a visual stimulus device in SSVEP-based BCI. In this study, SSVEP is measured using smart glasses and an LCD under almost identical conditions, and the classification accuracy of both are compared by training and classifying each SSVEP using a convolutional neural network (CNN). In the experiment, eight electrodes are placed around the subject's visual cortex and four different frequencies of flashing visual stimuli are simultaneously presented. EEGNet [2], a widely used deep learning model, is used to classify the EEG. The classification accuracy is compared for 0.5, 1.0, 1.5, 2.0, and 2.5 s of input data to EEGNet. Subsequently, these values are compared with those obtained using the LCD and are found to be approximately 6% less accurate than those obtained with the LCD.