非线性PCA在表面肌电信号特征提取中的应用

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.

Nonlinear Principal Component Analysis for Feature Extraction of SEMG

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. In this paper, 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 -- 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.