官金安,杨建华,赵瑞娟.基于ICA和极限学习机的模拟阅读脑电特征分类[J].中南民族大学学报自然科学版,2018,(1):85-89
基于ICA和极限学习机的模拟阅读脑电特征分类
EEG Feature Classification of Imitating-Reading Based on ICA and Extreme Learning Machine
  
DOI:
中文关键词: 模拟阅读  N2-P3成分  极限学习机
英文关键词: imitating-reading  N2-P3  ELM
基金项目:国家自然科学基金资助项目(91120017)
作者单位
官金安1,2,杨建华1,赵瑞娟2 1 中南民族大学 生物医学工程学院认知科学国家民委重点实验室 武汉 430074
2 中南民族大学 医学信息分析及肿瘤诊疗湖北省重点实验室武汉 430074 
摘要点击次数: 160
全文下载次数: 153
中文摘要:
      为了有效地提取N2-P3成分,利用ICA对脑电数据进行盲源分离,自动提取N2-P3成分;同时为了克服传统方法如支持向量机、神经网络训练时间长、个别识别准确率不够高的缺点,选择极限学习机作为分类器.在模拟阅读实验范式下,记录了7名受试者的脑电数据,利用ICA分别对每名受试者的高维脑电数据进行盲源分离,提取出 N2-P3成分,以此作为靶特征,并与非靶特征一起放入极限学习机分类器进行分类.训练得到7名受试者的训练时间和分类准确率,并与支持向量机进行了比较.结果表明:经过ICA特征提取后,使用极限学习机进行分类,该分类器学习速度快,泛化能力强,训练时间大大减少.在分类准确率上,ICA+ELM 的分类准确率较传统的最佳单通道+SVM 有较大幅度的提升,从后者平均的 82.4% 提升到了 97.7%.
英文摘要:
      In order to effectively extract the N2-P3 components, ICA was introduced; At the same time, Extreme Learning Machine(ELM) was used as the classifier. The EEG data of 7 subjects were recorded, the ICA was used to separate the high-dimensional EEG data from each subject, and the N2-P3 component was extracted as the target Characteristics, and with untargeted features into the ELM for classification. The training time and classification accuracy of the seven subjects were trained and compared with the SVM. The results show that, after the ICA feature extraction, the training time is greatly reduced. In the classificationaccuracy, Compared with SVM, classification accuracy of ICA + ELM has a more substantial increase, from the latter average of 82.4% to 97.7%.
查看全文   查看/发表评论  下载PDF阅读器
关闭