基于多信号特征和D―S证据理论融合下肢运动模式识别.docVIP

基于多信号特征和D―S证据理论融合下肢运动模式识别 　　摘 要：针对下肢运动模式识别中出现的单一信号特征识别方法不能准确、可靠地识别出下肢的运动模式，以及多信号特征识别方法所带来的维数灾难问题。提出一种利用神经网络特征级融合多信号特征以及D-S证据理论对预识别结果进行决策级融合的下肢运动模式识别方法。对采集到的下肢表面肌电信号以及髋关节信号进行特征提取，对两种来自不同信号源的特征向量分别建立各自的神经网络，利用神经网络的特征级融合属性得出不同信号源特征对下肢动作的预识别结果。最后利用D-S证据理论融合来自不同信号源特征对待识别动作的概率信度，以实现决策级融合的目的。通过文中方法对五种下肢常见动作的识别效果进行了验证。实验结果表明，文章方法相比于单一信号源特征的识别方法具有更高的可靠性和正确率。 　　关键词：运动模式；数据融合；多源信息；神经网络；D-S证据理论；特征提取 　　中图分类号：R318.04 文献标志码：A 文章编号：2095-2945（2018）01-0025-03 　　Abstract： In view of the single signal feature recognition method which is adopted in the lower limb motion pattern recognition and can not accurately and reliably identify the lower limb motion pattern and the dimension disaster caused by the multi-signal feature recognition method， a method of lower limb motion pattern recognition using neural network feature level fusion and D-S evidence theory for decision level fusion of pre-recognition results is proposed. The features of the lower limb surface muscle signals and hip joint signals were extracted， and the neural networks were established for the two kinds of feature vectors from different signal sources. The feature level fusion attribute of neural network is used to obtain the pre-recognition results of different signal source features to lower limb motion. Finally， D-S evidence theory is used to fuse the probability reliability of different signal sources dealing with the recognition action， in order to achieve the purpose of decision level fusion. The recognition effect of five common lower extremity movements is verified by the method of this paper. The experimental results show that the proposed method is more reliable and accurate than the single signal source recognition method. 　　Keywords： motion pattern； data fusion； multi-source information； neural network； D-S evidence theory； feature extraction 　　引言 　　佩戴假肢是大腿截肢患者恢?突?本行为能力的重要方式之一。截至2006年4月1日的统计数据表明，我国肢体残疾人的数量为2412万，下肢截肢人数约为158万人，占截肢总数的85%，其中能够安装假肢的约为44万人[1]。由于智能假肢能够识别穿戴者的运动意图，实现人-假肢-环境三者之间有效的信息交互与控制，目前受