Identification of emotions evoked by music via spatial-temporal transformer in multi-channel EEG signals.

Introduction: Emotion plays a vital role in understanding activities and associations. Due to being non-invasive, many experts have employed EEG signals as a reliable technique for emotion recognition. Identifying emotions from multi-channel EEG signals is evolving into a crucial task for diagnosing emotional disorders in neuroscience. One challenge with automated emotion recognition in EEG signals is to extract and select the discriminating features to classify different emotions accurately. Methods: In this study, we proposed a novel Transformer model for identifying emotions from multi-channel EEG signals. Note that we directly fed the raw EEG signal into the proposed Transformer, which aims at eliminating the issues caused by the local receptive fields in the convolutional neural networks. The presented deep learning model consists of two separate channels to address the spatial and temporal information in the EEG signals, respectively. Results: In the experiments, we first collected the EEG recordings from 20 subjects during listening to music. Experimental results of the proposed approach for binary emotion classification (positive and negative) and ternary emotion classification (positive, negative, and neutral) indicated the accuracy of 97.3 and 97.1%, respectively. We conducted comparison experiments on the same dataset using the proposed method and state-of-the-art techniques. Moreover, we achieved a promising outcome in comparison with these approaches. Discussion: Due to the performance of the proposed approach, it can be a potentially valuable instrument for human-computer interface system.