基于表面EMG功率谱和BP网络的多运动模式识别

Multi-pattern recognition by using surface electromyography′s power spectra and BP neural networks

结合功率谱比值法和BP神经网络提出一种基于表面肌电信号(EMG)的多运动模式识别算法.该算法首先根据表面肌电信号功率谱的特点,提出一种有效的特征提取算法——功率谱比值法;然后将功率谱比值特征作为BP神经网络的输入向量,实现对伸腕、屈腕、张开、合拢四种动作模式的识别,该识别结果可为肌电假手的多种运动模式提供仿生控制的信号源.实验结果表明,该方法对同一健康受试者四种运动模式的识别成功率平均达到95%,而对不同的健康受试者的识别成功率平均达到85%.

An algorithm based on surface EMG signals was proposed by the combination of power spectral coefficient with BP neural networks （BPNN）, to implement multi-pattern recognition of surface electromyography （SEMG）. An effective method of feature extraction, power spectral coefficient method, was introduced. Then, it took the obtained characteristics （namely, the computed power spectral coefficients） as the inputs of BPNN to discriminate four motion patterns, palmrs dorsiflexion, flexion, opening and closing. The recognition results could be used as source signals to control powered prostheses. The experimental results indicate that, for the same healthy testee, the success rate can reach 95% averagely by using the above algorithm to implement four motion-pattern discrimination, while, for different healthy testees, it can reach 85 % averagely.