Permuted layer-based CNN for Emotion Detection with Multi-Modality Physiological Signals

Human emotion detection is a challenging task. Emotions are expressed via facial expressions, body/face temperature, EEG recordings, voice recordings. In this paper, a novel CNN based model is reported to detect human emotion using multimodal human emotion representation data. The model is based on the concepts of one-dimensional convolution on the permuted-ensembled feature space representing EEG signals. The model addresses the issues with localized one-dimensional convolution over high-dimensional feature spaces and attempts to improve the classification performance. Results indicate improved detection accuracy over different emotions as compared to other popular machine learning approaches.