Comparison of GSR and PPG signals in Arousal and Valence

초록

Human bio-signals can be used in various fields such as aviation, vehicles, and systems, but short cycles of emotional recognition and bio-signals suitable for the situation must be selected. Among the representative bio-signals, the PPG signal includes changes in optical blood flow and regular heart information. However, the GSR signal consists of physical details such as sweat output and body temperature change information according to emotional changes. In this paper, the results are compared to derive signals suitable for recognizing physical and emotional changes. We also used PPG and GSR signals from the Asian multimodal dataset MERTI-Apps to compare the proposed method's results and experimental progress. The used data is subjected to essential preprocessing and window slicing in a short period of 1.1 seconds to remove noise from the signal. In addition, PPG and GSR signals were divided into three classes based on Arousal and Valence. The results of the 1D CNN (convolutional neural network) model were compared to derive signals suitable for recognizing each physical and emotional change. The experiment results were 53.78% based on GSR signal accuracy and 64.19% based on Arousal, and compared to the PPG signal, Arousal was about 1%, but the Valence was about 5%, which was lower than the PPG signal.