Automated emotion recognition plays a vital role in problem solving, decision making and social activities of human life. An emotion is a set of reactions and experience to a given conditions, which are modeled as a linear combination of arousal and valence dimensions. Emotional pattern recognition based on physiological signals is a relatively new and fast growing area of research. Several physiological signals such as ECG, EEG, and EMG have been used for emotion recognition. Analysis of Electrodermal Activity (EDA) signals is one of the popular technique for emotion state analysis. In this work, an attempt is made to discriminate arousal-valence dimensions using EDA signals and multiscale one dimensional convolution neural network (MCNN). For this, EDA signals are obtained from publically available online DEAP database. These signals are normalized using channel normalization and subjected to MCNN for event-related robust features and classification. K-fold cross validation is used to investigate the performance of classifier. The result shows that the MCNN are able to discriminate the emotional states in arousal/valence dimensions. The proposed approach obtained an overall classification accuracy of 83.75% and 81.25% for arousal and valence scale, respectively. The network yields better classification performance for arousal scale then valence diemension. This might be due to the fact that arousal represent the intensity of emotions. The result also show that the proposed approach is better than the conventional hand-crafted feature based approach. Thus, it appears that the proposed approach can be used to differentiate autonomic and clinical conditions.
Keywords: Emotion; classification; convolution neural network; deep learning; electrodermal activity.