针对肌电信号的非平稳特性 ,采用小波变换方法对表面肌电信号进行分析 ,提取小波系数最大值构造特征矢量输入神经网络分类器进行模式识别 ,经过训练能够成功地从掌长肌和肱桡肌采集的两道表面肌电信号中识别展拳、握拳、前臂内旋、前臂外旋四种运动模式。实验表明 ,基于小波变换的神经网络分类方法所需的数据短、运算快 ,对于肌电假肢的控制具有良好的应用前景。 Aiming at the non-stationary characteristics of EMG signal, the surface electromyography (EMG) signal was analyzed by wavelet transform, the maximum value of wavelet coefficients was extracted, and the feature vector was input to neural network classifier for pattern recognition. After training, Muscle acquisition of the two surface EMG signals identify boxing, fist, forearm pronation, forearm rotation four kinds of exercise patterns. Experiments show that the neural network classification method based on wavelet transform needs short data and fast computation, and has good application prospect for the control of EMG prosthesis.