VR环境下运动想象脑电分类算法及脑机交互应用  

Motor Imagery EEG Classification Algorithm and Brain-computer Interaction Application in VR Environment

作　　者：田杰铭 程时伟[1] TIAN Jie-ming;CHENG Shi-wei(School of Computer Science Zhejiang University of Technology,Hangzhou 310023,China)

机构地区：[1]浙江工业大学计算机科学与技术学院,杭州310023

出　　处：《小型微型计算机系统》2022年第10期2070-2077,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(62172368)资助;浙江省自然科学基金项目(LR22F020003)资助.

摘　　要：在虚拟现实环境下开发脑机交互应用,有助于提升被试的沉浸感,可用于提升肢体运动障碍患者的康复训练效果.但是已有的运动想象脑电分类算法准确率和稳定性较低,导致基于运动想象的脑机交互应用系统实用性不高.为此,本文对脑电信号先进行时序分片,提出基于2D CNN-LSTM的运动想象脑电分类算法,对脑电信号的时序特征与通道连通性进行充分提取.使用64导联的脑电仪采集被试在VR环境下的左右手运动想象脑电信号,结果发现本文方法相较于其他方法有着更好的稳定性,离线分析时个体模型在测试集上的平均准确率达到94.8%,平均F1-score达到0.951,优于其他算法,并存在显著性差异.此外,本文在世界脑控机器人大赛数据集上进行了测试,平均准确率达到了88.7%,优于所对比的其他算法.在线使用个体模型对VR下运动想象脑电进行实时分类,平均准确率达到了91.7%,相较于现有的VR-BCI有较为明显的提升.The development of brain-computer interaction applications in the virtual reality environment can help improve the immersion of the tested users,and can be used to improve the rehabilitation training effect of patients with limb movement disorders.However,the accuracy and stability of the existing motor imagery EEG classification algorithms are low,and the practicability of the brain-computer interaction application system based on motor imagery is not high.Therefore,First,the EEG signal is segmented in time series,and a 2D CNN-LSTM-based motor imagery EEG classification algorithm is proposed to fully extract the time series features and channel connectivity of EEG.The 64 channels EEG device was used to collect the EEG signals of the left and right hand motor imagination of the tested users in the VR environment.It was found that the algorithm in this paper has better stability than other methods.In offline analysis,the individual models are averaged on the test set.The accuracy rate reaches 94.8%,and the average F1-score reaches 0.951,which is better than other algorithms,and there is a significant difference.In addition,this paper is tested on the dataset of the World Brain Control Robot Competition,and the average accuracy rate reaches 88.7%,which is better than other algorithms compared.Using the individual model online to decode the MI EEG under VR in real time,the average accuracy rate has reached 91.7%,which is a significant improvement compared to the existing VR-BCI.

关 键 词：虚拟现实 脑机接口 脑电信号 卷积神经网络 长短期记忆网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]