非线性PCA在表面肌电信号特征提取中的应用

Nonlinear Principal Component Analysis for Feature Extraction of SEMG

针对表面肌电信号的特点,提出了一种应用非线性主分量分析(PCA)提取表面肌电信号特征的新方法.该方法在表面肌电信号滤波的基础上,采用非线性PCA方法完成数据压缩,将多路表面肌电信号转换为一维的特征数据主元,并以主元曲线的形式输出特征提取结果.本文采用基于自组织神经网络的非线性PCA对手臂尺侧腕伸肌和尺侧腕屈肌的两路表面肌电信号进行主元提取,试验结果表明,四种手部运动模式(握拳、展拳、腕外旋、腕内旋)对应的表面肌电信号利用该方法处理后,得到的主元曲线具有很好的类区分性,依据所得主元曲线的形状特征可以有效地进行手部动作类别的识别.

In connection with the character of Surface Electromyography signal (SEMG), a new method that uses nonlinear Principal Component Analysis (NLPCA) to extract feature from SEMG was proposed. After filtering SEMG, it utilizes NLPCA to achieve data compression, which transforms multi-way SEMG to one dimensional feature data saying principal component, and then,outputs the extraction in principal curve.NLPCA basing on auto-associative neural networks was utilized to extracted principal component from two-way SEMG, which derived from ulnar extensor muscle and ulnar flexor muscle of wrist respectively. Experimental results showed that, after processing SEMG of four hand motion patterns that including fist clenching, fist unfolding, wrist intorsion and wrist extortion with this method, principal curves with good character of category division were produced. According to the shape features of principal curves, motion of hand can be recognized efficiently.