基于改进人工蜂群优化与组合特征提取的手部运动意图识别

Hand movement intention recognition based on improved artificial bee colong optimization and combined feature extraction

为了解决表面肌电信号混迭导致的手部运动意图识别率较低的问题,提出了一种基于改进的人工蜂群优化盲源有序分离算法。本算法以表面肌电信号的规范四阶累积量作为代价函数,使用改进的人工蜂群优化算法代替传统的梯度算法对代价函数进行优化,并以代价函数绝对值的降序逐次提取出源信号;对于肌电信号的非平稳性及易受干扰的问题,采用一种基于小波包变换和样本熵的特征提取方法,并与表征肌电信号细节和强度的特征峰度、偏度、肌电积分值组合构建特征向量,训练二叉树支持向量机分类器。实验结果表明,采用表面肌电信号的盲源分离预处理与组合特征提取的方法识别六种手部运动意图,平均准确率达到93. 33%。

A sequential blind source signal separation algorithm based on Improved Artificial Bee Colony Optimization(IABO)algorithm was proposed to solve the problem of low recognition accuracy of hand movement intention,which caused by aliasing of surface electromyography(sEMG)signal.Normalized fourth-order cumulant of the sEMG signal was used as cost function in the algorithm,and traditional gradient algorithm was replaced by IABO for optimizing the cost function.Therefore,the source signal could be extracted on the descending order of normalized fourth-order cumulant;Considering the characteristics of non-stationary and easily disturbed of sEMG signal,we proposed a combined feature extraction method,which contained sample entropy from the subspace of wavelet packet decomposition and kurtosis、skewness、the integral values of sEMG signal.At last,binary tree support vector machine classifier with combined features was trained.The experimental results show that,six kinds of hand movements intention can be effectively predicted by using the blind source separation and combined feature,the average accuracy rate reaches to 93.33%.