基于特征融合神经网络的运动想象脑电分类算法

Motor imagery EEG classification algorithm based on feature fusion neural network

运动想象-脑机接口(MI-BCI)技术为运动障碍患者提供了一种新的与外界交流的能力。应用卷积神经网络(CNN)处理运动想象(MI)脑电分类问题时,多提取最后卷积层的特征,忽视了中间层大量可用信息,导致MI-BCI的分类性能较差。针对这一问题,提出模型内层融合(WMFF)和模型间层融合(CMFF)两种特征融合策略。WMFF策略提取CNN每一层特征进行融合;CMFF策略融合CNN和长短时记忆网络并提取每一层特征。本研究用BCI竞赛IV Datasets2a数据集对所提方法进行验证,WMFF和CMFF MI脑电信号四分类平均正确率分别达到76.19%和80.46%。结果表明,所提方法可有效提高MI脑电信号分类正确率,为MI脑电信号分类提供了新的思路。

Motor imagery-based brain computer interface(MI-BCI)technology enables patients with movement disorders to acquire a new ability to communicate with the outside world.However,when using convolutional neural network(CNN)for MI electroencephalogram(EEG)classification,researchers often extract the features of the final convolutional layer and ignore the large amount of available information in the middle layer,resulting in poor classification performance of MI-BCI.To solve this problem,two kinds of feature fusion strategies,namely with-in model fusion-feature(WMFF)and cross model fusion-feature(CMFF),are proposed.WMFF strategy extracts the features of each CNN layer separately for feature fusion;while CMFF strategy integrates CNN and long short-term memory network and extracts the features of each layer.BCI competition IV Datasets 2 a is used to verify the proposed method,and the results show that the average accuracies of WMFF and CMFF for 4-category MI EEG classification reach 76.19%and 80.46%,respectively,which indicates that the proposed method can effectively improve the accuracy of MI EEG classification,and provide new ideas and methods for the application of MI-BCI.