EEG-based emotion recognition using deep learning

It has gained importance in many areas to support communication, one of the most basic needs of today, and to make visible the extraordinary brightness of facial expressions. Simplifying emotions has important contact information for healthy or sick individuals. For people who have difficulty expressing emotions due to some health conditions they experience, emotion recognition is very important in their life capacities. Since emotional states can be detected by examining EEG treatment for emotion recognition of the waves occurring in the brain, emotion recognition is aimed using deep learning. The deep learning method was preferred because the classification made with machine learning methods was left behind in terms of classification performance with deep learning methods. In this thesis study, 8 (engineer: 4, architect: 4) participants studying at Konya Technical University Electrical - Electronics Engineering and Architecture departments took part. An experimental setup was created through the open source software package PsychoPy, and participants were shown images of these architectural structures for certain and equal periods of time to examine the effects of iconic and ordinary architectural structures on emotional states. While applying this experimental setup, EEG signals were recorded simultaneously in the EmotivPro program with the 14-channel Emotiv EPOC X wireless EEG headset. After applying a Butterworth filter to the recorded raw EEG signals, the signals were normalized. Then, time-frequency analyzes were applied separately to STFT and CWT architect and engineer participant groups. This data set, created with architectural stimulus images, was classified with the simple CNN model. As a result, a high classification (validation) accuracy of 91.67% was achieved in the architect participant group with the STFT analysis method. In the engineer participant group, it was classified with 89.52% accuracy by the STFT analysis method. Classification rates made with SDD analysis produced lower accuracy than classification rates made with KZFD. In this study, it was examined that architectural structures have an effect on emotional states, and high performance results were observed using EEG signals and deep learning.