基于深度学习的脑电信号情感识别研究进展

Research progress in emotion recognition of EEG signals based on deep learning

情感识别能有效促进智能人机交互的发展,辅助医学诊断,已成为人工智能领域的研究热点之一。脑电图(electroencephalogram,EEG)是对情感状态波动反应最为灵敏的信号,但传统的机器学习算法受到脑电信号非平稳性和个体差异性等特性的制约,很难进一步提高分类准确率和模型泛化能力。近年来能自动化特征提取的深度学习方法愈受学者青睐。本文对基于深度学习的脑电信号情感识别研究进行归纳总结,简述情感识别的相关理论以及常用公开数据集,总结对比不同情感识别模型和EEG传统特征、EEG原始数据以及多模态信号融合特征三类输入数据对分类精度的影响,最后探讨现阶段研究所存在的问题并展望该领域未来发展方向,以期能为后续研究提供借鉴。

Emotion recognition which can effectively promote the development of intelligent human-computer interaction and assist medical diagnosis has become a research hotspot in the field of artificial intelligence.Electroencephalogram(EEG)is the most sensitive signal to the fluctuation of emotional state,but traditional machine learning algorithms are difficult to further improve the classification accuracy and model generalization ability due to it’s non-stationary and individual difference.In recent years,deep learning methods that can automate feature extraction are more and more favored by scholars.This paper summarizes the research methods on EEG-based emotion recognition using deep learning algorithm.The related theories of emotion recognition and the public datasets which are in common use have been mentioned.The effects influence on classification accuracy of different emotion recognition models and input data which is divided into EEG traditional features,EEG raw data and multimodal signal fusion features have been summarized and compared in this article.Finally,this paper discusses the existing problems in current research and looking forward to the future development direction in this field,in order to bring reference for follow-up research.