基于脑电信号的情感识别方法综述

A survey of emotion recognition method based on EEG signals

情感识别是让计算机感知人类情感状态从而进行人机情感交互的关键技术,已经成为人工智能领域的研究热点。计算机对人类情感的感知可分为“感”和“知”两部分:“感”是指计算机对人类面部表情和语音等非生理信号以及外围神经和脑部电信号等生理信号的获取;“知”是指计算机对获取信号的识别和推断。基于脑电信号的感知方法因其具有较高的识别率而成为情感识别最主要的手段之一,其主要步骤为脑电信号获取、预处理、特征提取以及分类识别。本文对脑电情感识别方法中各步骤所涉及的研究方法进行了归纳和总结,介绍了脑电信号的采集和常用数据库以及去除伪迹信号的预处理方法,从时域、频域、时频域和非线性动力学角度归纳了脑电信号的特征提取方法,总结了常用的机器识别分类的无监督学习和有监督学习方法,最后探讨了脑电信号情感识别研究中存在的问题并展望了未来发展方向,以期为相关研究带来一定的借鉴。

Emotion recognition is a key technology that allows computers to perceive the emotional state of human beings and thus engage in human-computer emotional interaction.It has become a research hotspot in the field of artificial intelligence.The perception of human emotions by computers can be divided into two parts:“sense”and“knowledge”.“Sense”refers to the computer acquisition of non physiological signals such as human facial expressions and speech,and physiological signals such as peripheral nerve and brain electrical signals.“Knowledge”refers to the recognition and inference.Emotion recognition based on electroencephalogram(EEG)has become the main research method because of its high recognition rate and its main steps are EEG acquisition,preprocessing,feature extraction and classification recognition.In this paper,the research methods involved in each step of EEG emotion recognition are summarized.The collection of EEG signals,common databases and preprocessing methods of removing artifacts are introduced.The feature extraction methods of EEG are summarized from the perspectives of time domain,frequency domain,time-frequency domain and nonlinear dynamics,and the unsupervised learning and supervised learning methods of machine recognition classification are introduced.Finally,the paper discusses the existing problems in the research of emotion recognition of EEG signals and looks forward to the future development direction,in order to bring certain reference for the related research.