基于熵和PSO优化SVM的肌电信号跌倒识别  

Fall Recognition Based on EMG Signal Entropy and PSO-SVM

作　　者：武昊[1] 席旭刚[1] 罗志增[1] 

机构地区：[1]杭州电子科技大学智能控制与机器人研究所,杭州3100118

出　　处：《传感技术学报》2015年第11期1586-1590,共5页Chinese Journal of Sensors and Actuators

基　　金：国家自然科学基金项目(60903084;61172134);浙江省自然科学基金项目(LY13F030017);浙江省科技计划项目(2014C33105)

摘　　要：跌倒对老年人的健康构成严重危害,设计了一种基于肌电信号的跌倒识别方法,可用于跌倒检测报警。该方法首先对表面肌电动作信号进行小波包分解,再依据信号特征选取信号的低频分量并重构,计算其排列组合熵,最后以4路肌电信号对应的排列组合熵组成的特征向量输入SVM进行模式识别并采用粒子群算法对SVM中惩罚参数c和核函数参数g进行优化,对8种动作进行识别实验,跌倒识别率88%,特异度98.3%,平均识别率97.0%,优于网格法和遗传算法支持向量机(GASVM)的参数优化,具有较强的鲁棒性和抗干扰能力。A new fall detection method was designed for fall alarm based on s EMG. Firstly,the s EMG signals are de-composed into subspaces with wavelet packet. Then,depending on the signal characteristics,signals of low-frequen-cy component were recombined to calculate the permutation entropy. Finally,the SVM method was used to recog-nize eight actions according to the permutation entropy of four s EMG signals,and the particle swarm optimizationwas used to optimize punishment parameter c and nuclear parameter g. The result shows fall sensitivity,fall specificity,the average recognition rate were 88%,98.3%,97.0%,better than the gird method and genetic algorithm pa-rameters optimization. The method has strong robustness and noise immunity.

关 键 词：表面肌电信号 小波包分解 排列组合熵 支持向量机 粒子群算法 

分 类 号：TP212.3[自动化与计算机技术—检测技术与自动化装置]