EEG-based Incongruency Decoding in AR with sLDA, SVM, and EEGNet

Abstract

Augmented reality (AR) technologies enhance a user’s physical environment by providing contextual information about their surroundings. This information might appear incongruent to users, either due to their current mental context or factual errors in the data. This paper explores the feasibility of incongruency decoding using electroencephalographic (EEG) signals from 19 participants acquired during an interactive AR task. Previous studies on single-trial N400 decoding for brain-computer interfaces using EEG data are limited. Therefore, we implemented commonly used classification approaches and assessed their decoding performance compared to the convolutional neural network EEGNet. We found that the investigated approaches offer comparable accuracies ranging from 63.3% to 64.8%. Successful decoding of incongruency effects can foster more contextually appropriate interactions within AR environments.

Keywords

• electroencephalography (eeg)
• augmented reality (ar)
• brain-computer interface (bci)
• deep learning