手部抓取动作特征提取算法研究

发布时间：2018-10-24 16:10

【摘要】：针对人手抓取动作问题,如何有效地提取表面肌电信号特征是提高其模式识别率的关键。通过对前人不同手部抓取动作的分类方法及日常生活工作中使用的频度进行统计学分析,决定选取8种抓取动作进行研究。实验显示,随着手部动作姿态种类的增加,基于表面肌电信号的不同特征提取算法分类能力出现不同程度的下降甚至失效。为取得更为理想的抓取动作分类效果,提出将抓取动作分割为预抓取和抓取两个动作过程。选择采集预抓取动作前臂肌电信号,除对其时域、频域及时频域常用特征量进行分析对比外,还增添了对时频域中小波系数最大模值的分析,旨在找出最有效表征肌电信号动作分类的特征量。实验结果表明,小波系数最大模值量最有特征可分性,区分效果比较理想。
[Abstract]:How to extract the features of surface electromyography (EMG) effectively is the key to improve the pattern recognition rate. Based on the statistical analysis of the classification methods of different hand grabbing movements and the frequency used in daily life, it is decided to select 8 grab movements to study. Experiments show that the classification ability of different feature extraction algorithms based on surface EMG signals decreases or even fails with the increase of hand movements and postures. In order to achieve a more ideal classification effect of grab action, it is proposed that the grab action be divided into two processes: pregrab and snatch. Choosing to collect the forearm electromyography signals, besides analyzing and comparing the time-domain and frequency-domain characteristics, the maximum modulus value of time-frequency domain wavelet coefficients is also analyzed. The aim of this paper is to find out the characteristic quantity which is the most effective representation of EMG motion classification. The experimental results show that the maximum modulus value of wavelet coefficients has the most characteristic separability and the discrimination effect is ideal.
【作者单位】：哈尔滨理工大学自动化学院;黑龙江科技大学机械工程学院;
【基金】：国家”863”重大项目子课题高性能仿人型假手(2009AA043803) 哈尔滨市科技创新人才基金(2009RFQGG207) 黑龙江省研究生创新科研基金(YJSCX2009-059HLJ)
【分类号】：TN911.7

【相似文献】