Alternative classification techniques for brain-computer interfaces for smart sensor manufacturing environments
- Additional Document Info
- View All
© 2015, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved. Intelligent manufacturing requires new forms for sensing and controlling machines. Emerging type of controllers are the brain computer interfaces that detect intentions by using of brain signals and use them for control purposes. The common classification of EEG signals is complicated due to its variation over time and recordings. In this paper two classification algorithms are proposed for Brain Computer interfaces. The two algorithms are the Restricted Boltzmann Machines and the Long Short time Memory. The first one creates an internal representation of the joint entrance of the input and the output, while the second one uses its recurrent connection to keep an internal representation of previous inputs. The classification methods were tested using a Database of Brain activity patterns representing two classes (left and right motor imagery). Both techniques were compared against a conventional feedforward Artificial Neural Network trained with backpropagation, observing their classification accuracies. The classification accuracy obtained from the two proposed methods revised showed similar accuracy than for the common Artificial Neural Network, displaying their usability on classifying EEG patterns, and open the possibility to be used in manufacturing systems.