基于肌电信号的人手运动状态的辨识

研究的目的在于利用人体前臂的肌电信号进行人手动作模式的识别。根据采集的肌电信号，判断动作始末状态并对该肌电信号进行小波降噪预处理，利用小波变换的高频细节系数极值构造特征矢量，经过学习矢量量化（LVQ）神经网络训练，能够有效地识别握拳、展拳、手腕内旋和手腕外旋4种动作模式。和前馈型神经网络比较，LVQ神经网络具有更高的识别准确率和更稳定的再现性。

Motion State Identification of Human Hand Based on EMGs

The objective of this study is to identify the motion state of human hand based on the EMGs.Based on the raw EMGs,the starting and stopping states of the hand were first determined.The EMGs from the starting time to the stopping time were de-noised by wavelet analysis.Feature vectors were built from the maximum of high frequency coefficients of the wavelet transformation.A learning vector quantization(LVQ)network,whose input was the feature vectors,was used to classify four motion states of the hands.The four sates were hand grasping, hand opening,wrist inner spinning,and wrist outer spinning.The results showed that the correction rate and repeatability were higher than that with feedforward(FN) network.