基于HMM的表面肌电信号模式分类

Classification of surface electromyography patterns by using HMM

按等时间间隔将表面肌电信号(SEMG)划分为不同的段,利用小波变换对其进行特征提取,借助隐马尔可夫模型(HMM)的动态建模能力来感知不同动作模式下SEMG的时变特性.具体应用时,先根据样本对各动作模式下的HMM进行训练,待各模型参数稳定后,再利用HMM对特征提取后的SEMG进行模式分类.实验结果表明:该方法具有很好的分类识别率.在6个手部动作识别中,上翻、下翻、内旋和外旋4种动作的识别准确率均在90%以上.

Surface electromyography （SEMG） signal was divided into different segments with the same interval, whose feature was extracted by wavelet transform. The time-variable characteristics of SEMG signal under different motion patterns were apperceived by using the ability of hidden Markov model （HMM） in dynamic modeling. In actual application, the HMM under every motion pattern is firstly trained by samples. When the parameters of mode are stable, the HMM is applied to pattern classification of SEMG signal after feature extraction. The experimental results indicate that this method has high success rate in classification. In a six hand-motion recognition system, the success rates of four hand-motion patterns （namely, upwards,downwards, entad and forth） can reach over 90 %.