基于肌音信号分析的踝关节动作模式识别

Pattern Recognition Research on Ankle Action Modes Based on Mechanomyographic Signal Analysis

通过测量分析受试者小腿胫骨前肌和腓骨短肌的肌音信号,对踝关节背伸、跖屈、外展、内收等4个动作进行模式识别研究。提出了基于二次包络线的不等长信号分割算法,以及基于非线性小波变换的奇异值特征提取方法,并使用SVM分类器进行模式识别。结果表明:基于不等长分割的算法可以有效截取踝关节肌音信号的动作段信号;在两通道信号采集的情况下,利用非线性小波变换得到的奇异值特征在踝关节四模式识别中总体准确率可以达到87.8%,验证了本文提出的分析方法的有效性。

Two channel mechanomyographic signals were collected from anterior tibial muscle and peroneus brevis in the subjects＇ calves when their ankles acted dorsiflexion,plantar flexion,adduction and abduction,i.e.,four action modes.The collected signals were segmented by unequal division algorithm based on signals＇ second envelops and nonlinearly scaled wavelets were used to get singular-valuedecomposed（SVD）features from segmented signals,The SVM classifier was used to classify the action modes.Results showed that the second-envelop based unequal division algorithm can effectively intercept action segments from original signals and SVD features extracted from two channel signals by nonlinearly scaled wavelets can provide action modes recognition with the overall accuracy rate of 87.8%,which verified the feasibility of the proposed methods.