Li Jia Wen, Chen Rong Jun, Barma Shovan, Chen Fei, Pun Sio Hang, Mak Peng Un, Wang Lei Jun, Zeng Xian Xian, Ren Jin Chang, Zhao Hui Min
School of Computer Science, Guangdong Polytechnic Normal University, Guangzhou, 510665 China.
Guangxi Key Lab of Multi-source Information Mining & Security, Guangxi Normal University, Guilin, 541004 China.
Cognit Comput. 2022;14(6):2260-2273. doi: 10.1007/s12559-022-10053-z. Epub 2022 Aug 26.
Emotion can be influenced during self-isolation, and to avoid severe mood swings, emotional regulation is meaningful. To achieve this, efficiently recognizing emotion is a vital step, which can be realized by electroencephalography signals. Previously, inspired by the knowledge of sequencing in bioinformatics, a method termed brain rhythm sequencing that analyzes electroencephalography as the sequence consisting of the dominant rhythm has been proposed for seizure detection. In this work, with the help of similarity measure methods, the asymmetric features are extracted from the sequences generated by different channel data. After evaluating all asymmetric features for emotion recognition, the optimal feature that yields remarkable accuracy is identified. Therefore, the classification task can be accomplished through a small amount of channel data. From a music emotion recognition experiment and a public DEAP dataset, the classification accuracies of various test sets are approximately 80-85% when employing an optimal feature extracted from one pair of symmetrical channels. Such performances are impressive when using fewer resources is a concern. Further investigation revealed that emotion recognition shows strongly individual characteristics, so an appropriate solution is to include the subject-dependent properties. Compared to the existing works, this method benefits from the design of a portable emotion-aware device used during self-isolation, as fewer scalp sensors are needed. Hence, it would provide a novel way to realize emotional applications in the future.
在自我隔离期间，情绪可能会受到影响，为避免严重的情绪波动，情绪调节具有重要意义。要实现这一点，有效识别情绪是关键的一步，这可以通过脑电图信号来实现。此前，受生物信息学中序列知识的启发，一种名为脑节律序列分析的方法被提出用于癫痫检测，该方法将脑电图分析为由主导节律组成的序列。在这项工作中，借助相似性度量方法，从不同通道数据生成的序列中提取不对称特征。在评估所有用于情绪识别的不对称特征后，确定了具有显著准确率的最优特征。因此，可以通过少量通道数据完成分类任务。从音乐情绪识别实验和公开的DEAP数据集中可以看出，当采用从一对对称通道提取的最优特征时，各种测试集的分类准确率约为80%-85%。在关注资源使用较少的情况下，这样的性能令人印象深刻。进一步的研究表明，情绪识别具有很强的个体特征，因此合适的解决方案是纳入个体依赖属性。与现有工作相比，该方法受益于在自我隔离期间使用的便携式情绪感知设备的设计，因为所需的头皮传感器较少。因此，它将为未来实现情绪应用提供一种新方法。
