基于多尺度主元分析的表面肌电信号模式分类

Pattern Recognition of Surface Electromygram Based on Multi-scale Principal Component Analysis

用基于小波变换的多尺度主元分析提取表面肌电信号特征,然后用贝叶斯分类器进行模式分类。实验结果显示,当选用Harr小波和bior2.6小波对肌电信号进行5层小波分解时,该方法对前臂6种动作模式(内翻,外翻,握拳,展拳,上切和下切)的正确识别率可以达到99.44%。研究表明,该方法优于基于小波系数统计特征和主元分析降维相结合的特征提取方法,能成功识别出多种动作模式。

Multi-scale principal component analysis based on wavelet transform was applied in feature extraction ot sEMG, and bayes classifier was used for pattern classification in this paper. The experiment showed that when Harr wavelet or bior2.6 wavelet was employed to decompose EMG at 5 levels, this method resulted in good performance in the pattern recognition of six movements including varus, ectropion, hand grasps, hand extension, upwards flexion and downwards flexion, with the accuracy of 99.44 %. It was superior to the feature extraction based on the statistic feature of wavelet coefficients combined with dimension-reduce by PCA. The research indicated that the proposed method can successfully identify many kinds of movements.