基于GGA-Elman网络的头部体态语言sEMG识别

PDF下载HTML

Title:

An sEMG approach to recognize the body language of the head based on the GGA-Elman network

作者:

杨钟亮¹, 陈育苗²

1. 东华大学 机械工程学院, 上海 201620;
2. 东华大学 服装·艺术设计学院, 上海 200051

Author(s):

YANG Zhongliang¹, CHEN Yumiao²

1. College of Mechanical Engineering, Donghua University, Shanghai 201620, China;
2. Fashion · Art Design Institute, Donghua University, Shanghai 200051, China

关键词:

头部运动; 体态语言; 肌电; 肌肉; 时域分析; 神经网络; 遗传算法; 模式识别

Keywords:

head movement; body language; surface electromyography; muscle; time domain analysis; neural network; genetic algorithm; pattern recognition

分类号:

TP391

DOI:

10.3969/j.issn.1673-4785.201310047

摘要:

为提高头部体态语言表达"同意"与"不同意"态度的识别效果, 提出结合贪心遗传算法和Elman神经网络的表面肌电识别方法。通过前导实验分别采集8名被试者点头与摇头时颈部肌肉的表面肌电信号, 利用Wilcoxon秩和检验提取具有显著性差异的10个肌电时域特征值, 进而基于贪心遗传算法优化的Elman神经网络建立体态语言识别模型。实验结果表明, 该模型能成功识别自发表达"同意"与"不同意"的头部体态语言, 与标准Elman神经网络和BP神经网络的识别模型相比, 相关系数更高、均方误差更小, 对测试集的正确识别率提高了3.2%以上, 从而验证了该方法的可靠性。

Abstract:

In order to improve the recognition effects of the "agreement" and "disagreement" attitudes expressed by the body language of the head movements, a surface electromyography (sEMG) approach in combination with the greedy genetic algorithm (GGA) and the Elman neural network is proposed. The sEMG signals of the neck muscles were detected while eight participants were nodding and shaking their heads respectively during a pilot experiment. By means of the Wilcoxon’s signed-rank test, ten features of the sEMG time domain indices were extracted with significant differences. Furthermore, the body language recognition model was constructed based on the Elman network optimized by GGA. Experimental results show that the model can successfully recognize the "agreement and disagreement" attitudes spontaneously expressed by the different body languages of the head. Compared with the recognition models using the standard Elman and BP network, the correlation coefficient of this present model is higher, the mean squared error is less, and the correct recognition rate of the test set is increases by over 3.2%, which demonstrate the reliability of this approach.