[5K1-OS-48] OS-48

9:30 AM - 9:45 AM JST(12:30 AM - 12:45 AM UTC)

[5K1-OS-48-03] Emotion Recognition Using Depthwise Separable TCN Based on Single-Channel Photoplethysmography

〇Wenkai Hu¹, Kana Eguchi¹, Shota Kato¹, Manabu Kano¹ (1. Kyoto University)

Keywords:

Emoiton Recognition,Machine Learning,Biological Signal Processing

Emotion recognition has attracted increasing attention as an important research topic in the field of affective computing. Single-channel photoplethysmography (PPG) signals are valuable for inferring affective states because they are acquired non-invasively and directly reflect cardiovascular dynamics. This study aims to develop an emotion recognition model based on single-channel PPG signals. In particular, we seek to exploit integratively the temporal- and frequency-domain information inherent in PPG signals, which existing methods relying solely on single-channel PPG representations cannot sufficiently capture. Raw PPG signals and frequency-domain features extracted from the discrete wavelet transform (DWT) coefficients were used as inputs to the model, and learning was formulated as a binary classification problem between pleasant and unpleasant emotions based on valence and arousal labels. Within this framework, the proposed model can estimate pleasant and unpleasant emotional states from single-channel PPG signals. To efficiently process these time-series features, a temporal convolutional network (TCN) composed of depthwise separable convolutions is utilized. Experimental results showed that our method improves emotion recognition performance over conventional approaches.