基于多层时变功能脑网络特征的运动想象识别  

Motor imagination recognition based on multilayer time-varying brain functional network features

作　　者：罗志增[1] 郑文涛 LUO Zhizeng;ZHENG Wentao(Intelligent Control and Robot Research Institute,Hangzhou Dianzi University,Hangzhou 310018,China)

机构地区：[1]杭州电子科技大学智能控制与机器人研究所,浙江杭州310018

出　　处：《华中科技大学学报（自然科学版）》2024年第5期56-63,共8页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(62171171);浙江省自然科学基金重点资助项目(LZ23F030005).

摘　　要：针对运动想象脑电信号(EEG)识别中信号随时间的结构动态变化与网络分离整合过程被忽视等问题,提出一种基于多层时变功能脑网络的运动想象特征提取方法.本方法截取运动想象有效片段投入EEGLAB进行信号预处理;依据滑动窗口方法,设定合适长度与步长,将信号分成连续且部分重叠的时间窗口,将时间窗口截获的脑电数据生成多个脑网络,以节点间锁相值构建多层时变网络模型.首先通过多层时变网络不同层的网络拓扑分析与层间相似度量指标自适应确定其中核心网络层,提取其节点度和聚类系数用以描述网络空间功能连接;然后结合多层参与系数和多层聚类系数,描述脑电信号网络动态变化与分离整合特征,并组合两者成为多层时变脑功能网络特征向量,完成运动想象识别任务.用支持向量机识别的结果表明:基于所构建的网络特征向量分类准确率高达89.14%,高出对比所用的单层网络特征6.61%.Aiming at the problem that the dynamic change of signal structure with time and the separation and integration process of network were ignored in the recognition of motor imagination electroencephalography(EEG)signals,a method of feature extraction of motor imagination based on multi-layer time-varying functional brain network was proposed.In this method,the effective fragments of motor imagination were extracted and put into EEGLAB for signal preprocessing.According to the sliding window method,appropriate length and step length were set,and the signals were divided into continuous and partially overlapping time windows.The EEG data intercepted by the time window were generated into multiple brain networks,and the multi-layer timevarying network model was constructed based on the phase-locking values between nodes.First,the core network layer was determined through the network topology analysis of different layers of the multi-layer time-varying network and the inter-layer similarity metric,and the node degree and clustering coefficient were extracted to describe the functional connection in the network space.Then,multi-layer participation coefficient and multi-layer clustering coefficient were combined to describe the dynamic changes and separation and integration characteristics of EEG networks,and the two characteristices were combined to form the feature vector of multi-layer time-varying brain functional network to complete the task of motor imagination recognition.Results of support vector machine(SVM)identification show that the classification accuracy of the proposed network feature vector is as high as 89.14%,which is 6.61%higher than that of the single layer network feature used for comparison.

关 键 词：运动想象 脑电信号(EEG) 多层网络 特征提取 网络拓扑 

分 类 号：TH89[机械工程—仪器科学与技术]