【摘要】 随着肌音信号（MMG）的发展和模式识别技术的进步，基于肌音信号的人机交互技术成为可能。在信号的采集中，希望使用较低的采样率降低采集成本和减少处理时间，并使用高通滤波器减少运动伪影，但两者对识别率结果的影响有待展开研究。通过采集7名受试者手部8个动作的肌音信号，使用指数加权法分割信号段后，提取4个常见的时域特征，通过机器学习算法建立分类器。分别对原始信号重采样以改变采样率并对信号使用不同的截止频率，创新性地提出了研究肌音信号的采样率与高通截止频率对识别率影响的方法。与1 000 Hz的采样率相比，500 Hz采样率的识别精度下降不到1%,300 Hz采样率的识别精度下降不到2%；与高通截止频率相比，2 Hz与5 Hz的截止频率可以保留更多的信息频段，得到更高的识别率。结果表明适当降低采样率和高通截止频率可以有效地平衡识别精度和成本。更多还原

【Abstract】 With the development of mechanomyography(MMG)signal and the progress of pattern recognition technology,it is expected to realize the human-computer interaction technology based on MMG. In the process of the signal acquisition,it is hoped to utilize lower sampling frequency to reduce acquisition cost and shorten processing time,and adopt a high-pass filter to reduce motion artifacts. However,the impact of the two on the recognition rate results needs to be studied. In this paper,on the basis of collecting the MMG signal of 8 hand movements of 7 subjects,the exponential weighting method is used to segment the signal segments to extract 4 common time-domain features,and then a classifier is built by the machine learning algorithm. The original signal is subjected to re-sampling to change the sampling frequency and different cut-off frequencies are used for the signal. An innovative method is proposed to study the effect of the sampling frequency and the high-pass cut-off frequency of the MMG signal on the recognition rate. It is concluded that,in comparison with that of the sampling frequency at 1 000 Hz,the recognition accuracy of the sampling frequency at 500 Hz is reduced by less than 1%,and that at 300 Hz is reduced by less than 2%. In comparison with the high-pass cut-off frequency,when the cut-off frequencies are at 2 Hz and 5 Hz respectively,more information frequency bands can be retained and higher recognition rate can be obtained. The results show that reducing the sampling frequency and the high-pass cut-off frequency properly can balance the recognition accuracy and cost effectively.更多还原