改进BP神经网络的EMG手指运动识别

Optimized BP Neural Networks for EMG Finger Movement Recognition

在基于肌电信号(EMG)手指运动的模式识别中,稳定性和识别率是两个主要问题,为此提出了一种新的EMG模式识别算法。该算法采用现代信号处理理论中的AR模型和改进的BP神经网络相结合的算法,有效的解决了BP网络识别中落入局部极值问题。进行试验,将提取到的特征值输入MATLAB建立一个改进多层BP神经网络,识别三个不同类型的手指运动。实验表明,改进BP算法较传统BP算法获得了更高的识别精度,达到94%左右。

In the pattern recognition of Finger movement based on electromyography （EMG）, the Stability and Recognition rate are both the problem. The paper proposes a new method of pattern recognition of EMG signal. The method combination of the algorithm using BP neural network AR model and the improvement of modern signal pro-cessing in the theory of the algorithm, can effectively solve the problem of BP network into local extremum recogni-tion. To make the classification of the eigenvalues of the EMG, these eigenvalues have been inputted to the MAT-LAB to build up a improved multilayer BP neural networks. For the recognition of three different kinds of finger mo-tion＆#39;s EMG signals, the experiments show that the improved BP algorithm, to obtain higher recognition accuracy than the traditional BP algorithm, to around 94%.