CNN Based Emotion Regression Using EEG Signals

Abstract

Emotions are considered to be a characteristic of humans and play an important role in their everyday interactions. A distinctive trait that separates humans from computers is their unique emotions. Emotion recognition of humans by computers offers a variety of applications in the area of human-computer interaction and brain-computer interface. Regarding the importance of this area, this study investigates two-dimensional (2D) convolutional neural network (CNN) based emotion recognition by using electroencephalogram (EEG) signals. In doing so, the signals are firstly transformed into 2D images through the use of continuous wavelet transforms (CWT) and three different CNN models are then compared for emotion regression. The study proposed method is evaluated by using the DEAP database and via the EEG signals recorded from 32 subjects who were watching 40 music video clips with a length of 60s. Findings from the evaluation of the three models indicate that the accuracy of 97/90%, was achieved by the first proposed model which has the best performance. It is also observed that the performance of this model has considerably improved as compared with the results of other studies using the same method.