基于加速度的小波能量特征及样本熵组合的步态分类算法  

Gait Pattern Classification with Wavelet Energy and Sample Entropy Based on Acceleration Signals

作　　者：邢秀玉[1,2] 刘鸿宇[2] 黄武[3] 

机构地区：[1]中北大学信息与通信工程学院,太原030051 [2]中北大学运动医学研究所,太原030051 [3]四川大学计算机学院,成都610065

出　　处：《传感技术学报》2013年第4期545-549,共5页Chinese Journal of Sensors and Actuators

摘　　要：针对传统的使用小波分解系数作为特征对走路、上楼、下楼进行分类时不能对具有相同强度加速度信号的步态样本进行分类的问题,提出了一种基于样本熵和小波能量相结合作为特征的分类算法。利用三轴加速度传感器采集走路、上楼、下楼3种步态下的上臂加速度信号,将信号进行小波分解,提取能量特征和样本熵特征,构建决策树分类器和贝叶斯分类器。决策树分类器和贝叶斯分类器的总体分类精度分别为75%和78.75%,使用样本熵与小波能量作为特征的分类精度比仅使用小波能量的分类精度提高了15.85%和19.17%。就步态分类精度而言,样本熵与小波能量相结合的方法优于仅使用传统小波能量方法。It is unavailable for the classification of gait samples with a same-intensity acceleration signal by use of the traditional wavelet decomposition for three gait patterns（walking,up stair and down stair）,so a new method of sample entropy combined with wavelet energy was proposed.The three kinds of acceleration signals of one ＇ s upper limb were captured,then the characters of energy and sample entropy were measured for the construction of decision tree and Bayes classifier after wavelet decomposition of these signals were made.The general classification accuracy of decision tree and Bayes classifier were up to 75% and 78.75% which improved 15.85% and 19.17% compared with wavelet energy alone.As far as the precision of gait classification is concerned,the method based on sample entropy and wavelet energy was better than the wavelet energy alone.

关 键 词：步态分类 样本熵 小波能量 贝叶斯 决策树 计步器 加速度 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置]