Abstract:

A novel classifier that uses wavelet transform and Lyapunov exponent to construct feature vectors is presented.Using the advantage of back propagation in area of non-linear modeling，construct neural network to classify.It is efficient to classify motion states of the hands which include hand grasping，hand opening，wrist inner spinning and wrist outer spinning through disposal of four routes SEMG gathering from antbrachium extensor，flexor and pronator.The result indicates that the classifier has more higher recognize exactitude and stable reappearance.

摘要： 提出了一种新的模式分类方法，该分类法采用小波变换和李雅普诺夫指数构造特征矢量，利用Elman神经网络在非线性建模方面的优势，构建前馈神经网络，以此进行特征分类。通过对前臂伸肌、屈肌以及旋前肌采集的肌电信号的处理分析，有效地实现了对握拳、展拳、手腕内旋和手腕外旋4种动作模式的识别。结果表明该分类器有较高的识别准确率和更稳定的再现性。