Application of Deep Neural Networks for EEG Signal Processing in Brain-Controlled Wheeled Robotic Platform

EasyChair Preprint 9773

Abstract

The paper describes the nature of the operation neurocomputer interface, and provides its own prototype neurocomputer management system that includes a helmet with a platform Open BCI Cyton, BCI server based on PC and, indeed, wheeled robot with an onboard computer Raspberry Pi. The transmission of 16- channel EEG recordings registered by Open BCI Cyton to ALL server is carried out via Bluetooth protocol, and the Wi-Fi standard is used for communication between the robot and BCI server. The main task was to create and investigate the possibility of using deep learning technologies to classify the filtered signals corresponding to the frequency band of Alpha waves of encephalograms. The software architecture and algorithm of system functioning are presented, and also the convolutional neural network and a multilayer perceptron as a neuroclassifier are investigated. Filtering of EEG signals and their classification are performed on BCI servers. The neuroclassifier based on the convoluted network showed higher accuracy, but requires more computing resources for its implementation.

Keyphrases: Convolutional Neural Networks, EEG signals, Multilayer Perceptron, deep learning, “brain-computer” interface

@booklet{EasyChair:9773,
  author    = {Serhii Artemuk and Vitalii Brydinskyi and Ihor Mykytyn and Yuriy Khoma},
  title     = {Application of Deep Neural Networks for EEG Signal Processing in Brain-Controlled Wheeled Robotic Platform},
  howpublished = {EasyChair Preprint 9773},
  year      = {EasyChair, 2023}}