Emotion Recognition using Deep Convolutional Neural Network on Temporal Representations of Physiological Signals

Abstract

Human emotion recognition using Electroencephalography (EEG) signals has become paramount in the field of affective computing. These signals received considerable recognition from researchers, as it delivers low cost and simple results for recognizing emotions manually as well as automatic. In this research paper, we present a novel approach for emotion classification based on electroencephalographic signals taken from the AMIGOS dataset. The proposed work extract features 14 channels of EEG signals using a variant of a deep convolutional neural network on spectrogram images of signals containing time and frequency. The proposed model outperforms the previous deep learning models. Accuracy of arousal comes out to be 0.875 with an f1 score of 0.75789 while the accuracy of valence comes out 0.750 with an f1 score of 0.49693.