University of Twente Student Theses


MEG-driven emotion classification using convolutional and recurrent Neural Networks

Orlinski, Jakub (2020) MEG-driven emotion classification using convolutional and recurrent Neural Networks.

[img] PDF
Abstract:In this paper, a new multi-channel emotion classification method based on the novel magnetoencephalography (MEG) dataset CiNet is proposed. This paper falls into the field of Brain-Computer Interface (BCI) research, as it uses brain activity data for recognizing human emotions. It should prove a valuable contribution and a comparison, as most BCI research uses electroencephalography (EEG) data instead, primarily from the DEAP dataset. Using a combination of a Convolutional Neural Network (CNN) and a Recurrent Neural Network (RNN), the system will analyze the high-fidelity data in an attempt to recognize the emotional state of the subject. The CNN encodes spatial information, while the RNN tracks changes over time. Each part is evaluated separately as well as in conjunction, so as to establish the contribution of each of the aspects of analysis. Those model variations are evaluated on both raw MEG signals and the Power Spectrum Density (PSD) extracted from the signal. The experimental results show that the best model is the CNN+RNN combination trained on raw signal data, and it achieves a mean accuracy of 56.5% on the valence/arousal classification task.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science BSc (56964)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page