Classification of steady-state visual evoked potential signals using a dense convolutional neural network for brain-computer interface

Brain-computer interface (BCI) is a communication system in which user commands are transmitted to the outside world without involving the natural exit routes of surrounding nerves and muscles. BCI is especially important for users with reduced mobility such as the disabled. However, programs are being developed for a wide range of users to continue activities in the fields of safety, security and entertainment. In non-invasive BCIs, electroencephalography (EEG) is usually used due to its high resolution, ease of acquisition and cost-effectiveness in comparison with other brain activity monitoring methods. BCI based on SSVEP can automatically identify user commands through a series of signal processing steps including pre-processing, interference detection or correction, feature extraction and feature classification. BCI performance is usually evaluated in terms of classification accuracy, classification speed and number of available choices. One of the upcoming challenges is the need for a large amount of data for feature extraction, so the use of convolutional neural network as a solution to select the best features and automatically extract it works well even in small data. In this research, experiments were conducted on two SSVEP data sets using EEGNET and the classification results were compared with common methods such as CCA, LDA, and SVM, and the accuracy was ۸۶.۶% for the SSVEP-EXOSKELETON data set and ۶۹.۲% for the data set MASAKI NAKANISHI was obtained.BCI is an artificial intelligence system that can reveal a specific set of patterns in brain signals during five consecutive steps, which are: signal acquisition, pre-processing or signal amplification, feature extraction, classification and control interface.

Keywords:

Brain Computer Interface , SSVEP , convolutional Neural Networks