摘要
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,提取小波系数幅值的最大和最小值构造特征向量,输入BP神经网络可进行模式识别,网络经过学习能够成功地从表面肌电信号中识别展拳、握拳、前臂内旋、前臂外旋4种运动模式。比较了标准的BP算法和用贝叶斯正则化与Levenberg-Marquardt算法相结合的改进BP网络训练的结果。实验表明,改进的BP网络在训练速度和识别精度上都比标准的BP算法有了很大提高,这对于肌电假肢的控制具有良好的应用前景。
The application of improved BP neural network together with the wavelet transform to the classification of surface EMG signal is described. The data reduction and preprocessing of the signal are performed by wavelet transform. The network can identify such four kinds of forearm movements with a high accuracy as hand extension, clench fist, forearm pronation and forearm supination. This paper compares the results by standard BP algorithm with that of Bayesian regularization together with LM algorithm. Experimental result shows that the improved BP neural network has a great potential when applied to electromechanical prosthesis control because of its enhanced training speed and identification accuracy.
出处
《医疗卫生装备》
CAS
2005年第12期17-19,共3页
Chinese Medical Equipment Journal
基金
国家973项目(2005CB724303)资助。