【作者】 梁光金； 曹佃国； 刘梦梦；

【Author】 Guang-Jin Liang;Dian-Guo Cao;Meng-Meng Liu;Qufu Normal University;

【机构】 曲阜师范大学；

【摘要】 针对目前脑卒中引发的上肢瘫痪患者在康复训练中参与度不高,以及单纯依靠表面肌电信号进行康复训练的局限性,设计一种基于脑电(Electro-encephalogram,EEG)和表面肌电(Surface electromyogram,sEMG)协同处理的上肢动作识别模式。选取5个上肢动作,使用Delsys肌电采集设备采集8通道的肌电数据,使用BP脑电采集设备采集32通道的脑电数据。对信号进行预处理后分别提取sEMG的时、频域特征和EEG的熵特征。提出一种基于Fisher准则构建融合系数矩阵的特征融合算法。使用交叉熵损失函数作为粒子群优化(Particle swarm optimization,PSO)的粒子适应度函数,构建一种新的PSOC-SVM分类器。实验表明:PSOC-SVM分类器平均模型训练的时间由原来的60s缩短到20s减少了近40s,分类器对融合后的特征向量进行分类,识别率达到98.38%。更多还原

【Abstract】 A upper-limb movement recognition mode based on electroencephalogram(EEG) and surface electromyogram signal(sEMG) is proposed. This mode can meet the patients with upper-limb paralysis caused by stroke are not very involved in rehabilitation training and the limitations of rehabilitation training based sEMG. Select 5 upper limb movements, use Delsys sEMG acquisition equipment to collect 8-channel sEMG data, and use BP EEG acquisition equipment to collect 32-channel EEG data. The signal is preprocessed and then the time and frequency domain features of sEMG and the entropy feature of EEG are extracted respectively. A feature fusion algorithm based on Fisher criterion is proposed, which constructs the fusion coefficient matrix of feature. A new PSOCSVM classifier is proposed, which uses the cross-entropy loss function as the particle fitness function of PSO. Experiments show that the average model training time of the PSOC-SVM classifier is shortened from 60s to 20s, which is reduced by nearly 40s. The classifier classifies the fused feature vectors, and the recognition rate reaches 98.38%.更多还原