基于改进卷积神经网络的脑电信号焦虑情绪量化识别

Quantitative Recognition of Anxiety in EEG Based on Modified Convolutional Neural Network

关键词:  卷积神经网络  脑电信号  焦虑情感量化识别  非平稳时变信号处理  类别不平衡

Key words:  convolutional neural networks  EEG signal  quantitative recognition of anxiety emotion  nonstationary time-varying signal processing  class imbalance

摘要:

精确量化检出大学生的焦虑情绪并对病理因素进行追溯分析，是临床心理治疗和心理危机干预的重要环节，而基于脑电(Electroencephalograph,EEG)信号的深度学习是当前最具发展潜力的一种诊断方法。本研究对传统卷积神经网络(Convolutional Neural Networks,CNN)进行改进，提出并构造一个基于“扩展信息输入空间”的神经网络(Neural Network Based on Extended Information Input Space，NN-EIIS)模型，取代CNN末端的分类器；并引入具有独立性的被试对象焦虑量表得分(Score of Anxiety Scale，SAS)，作为焦虑情绪量化标准和训练样本集的输出。以某高校大学生为研究对象进行实验，结果表明所提出的方案不仅实现了对焦虑情感的精确量化识别，还能利用所得模型，在一定程度上对大学生焦虑障碍患者的某些重要的内在病理因素进行追溯分析。

Abstract:

Accurate quantification of college students' anxiety and retrospective analysis of pathological factors are important links in clinical psychotherapy and psychological crisis intervention.And the deep learning based on electroencephalograph (EEG) is one of the most potential diagnostic methods.In this study,the traditional convolutional neural networks (CNN) are modified,and a neural network based on extended information input space (NN-EIIS) model is proposed and constructed to replace the classifier at the end of CNN.Score of Anxiety Scale (SAS) of independent subjects was also introduced as the output of anxiety quantification standard and training sample set.Taking college students in a university as the research object,the results show that the proposed scheme not only realizes the accurate quantitative recognition of anxiety emotion,but also uses the obtained model to trace and analyze some important internal pathological factors of college students with anxiety disorder to a certain extent.