基于幅值立方和BP神经网络的表面肌电信号特征提取算法  

Feature Extraction Method Based on Amplitude Cubic and BP Neural Network of Surface ElectroMyoGraph(SEMG) Signals

作　　者：黄鹏程[1] 杨庆华[1] 鲍官军[1] 张立彬[1] 

机构地区：[1]浙江工业大学特种装备制造与先进加工技术教育部/浙江省重点实验室,杭州310032

出　　处：《中国机械工程》2012年第11期1332-1336,共5页China Mechanical Engineering

基　　金：国家高技术研究发展计划(863计划)资助项目(2009AA04Z209);国家自然科学基金资助项目(51075363);浙江省自然科学基金杰出青年团队资助项目(R1090674)

摘　　要：传统特征向量提取算法得到的特征向量无法识别手指不同角度动作,将幅值立方法引入特征向量的提取算法之中,并对立方计算后信号的特征量进行降数量级处理。实验表明,提出的特征向量提取算法对表面肌电信号微小特征差异的零偏差识别率达到75%,且大偏差保持在5%以下。设计的基于BP神经网络的手指运动模式分类器,能有效地提高手指运动形式的正确识别率。Feature vector obtained from traditional characteristic vector extraction algorithm cannot identify the different angle motion of the finger,this paper introduced a method of amplitude cubic to the characteristic vector extraction algorithm,and then decreased the orders of magnitude.The comparison with traditional feature vector identification algorithms suggests that the zero error rate in identification to the tiny different SEMG feature of the feature vector extraction algorithms proposed reaches more than 75%,meanwhile the huge error rate remained below 5%.And the finger movement pattern recognizer based on BP neural network was designed to enhance the correct rate of the mode of finger movement.

关 键 词：幅值立方法 神经网络 特征向量 偏差率 

分 类 号：TP241[自动化与计算机技术—检测技术与自动化装置]