融合频带能量特征和双向门控循环单元的运动想象意图识别  

Intention Recognition of Motor Imagery Combined Multiband Power Features and Bidirectional Gated Recurrent Unit

作　　者：韩向可 郭士杰[1,3] HAN Xiang-ke;GUO Shi-jie(School of Mechanical Engineering,Hebei University of Technology,Tianjin 300130,China;School of Mechanical Engineering,Anyang Institute of Technology,Anyang 455000,China;Hebei Key Laboratory of Smart Sensing and Human-Robot Interaction,Hebei University of Technology,Tianjin 300132,China)

机构地区：[1]河北工业大学机械工程学院,天津300130 [2]安阳工学院机械工程学院,安阳455000 [3]河北省机器人感知与人机融合重点实验室,天津300132

出　　处：《科学技术与工程》2020年第9期3662-3667,共6页Science Technology and Engineering

基　　金：国家重点研发计划(2016YFE0128700)。

摘　　要：针对脑电信号分类正确率低的问题,结合频带能量、小波包变换和双向门控循环网络,提出了一种基于频带能量特征序列和深度学习算法的运动想象意图识别方法。首先,利用小波包变换对脑电信号进行分解、重构,获得运动想象相关频带信号;其次,对所得频带信号进行加窗,并滑动截取,通过计算所截每段信号能量,实现能量特征的时序化分解;最后利用双向门控循环网络对脑电信号进行识别并输出分类结果。实验结果表明:所提算法取得了92.1%的分类正确率,表明所提方法是切实可行的,能够有效改善分类识别率。Due to the low classification accuracy of EEG signal, a novel method based on band power time-series and deep learning algorithm was proposed, which fused wavelet packet transform, band power features and bidirectional gated recurrent unit(BiGRU). Firstly, the band signals related to motor imagery, were obtained by using wavelet packet transformation to decomposition and reconstruction the raw EEG data. Secondly, the power of each EEG epoch cropped by a sliding time-window was calculated to achieve the sequenced feature vectors. Finally, a BiGRU model was used to classify the EEG signals and output the classification results. Experimental results show that the accuracy rate reaches 92.1% by using the proposed algorithm, which indicates the developed novel method is feasible effective for BCI system.

关 键 词：频带能量 双向门控循环单元 运动想象 脑机接口 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]