基于肌电信号的手臂运动状态的辨识

Motion State Identification of Human Elbow Joint Based on EMGs

本研究的目的是利用人体上肢肌肉的肌电信号辨识人体肘关节运动状态。当人体手臂做屈伸运动时,采集肱二头肌和肱三头肌的肌电(EMGs)信号和肘关节角度信号,对EMGs进行处理和特征提取。提取的特征值作为一个四层的神经网络模型的输入信号,运用改进后的误差反传学习算法最优化网络各层权值,映射出人体表面肌电信号和手臂运动状态间的非线性关系,并将处理后的肌电信号转换为相应时刻的肘关节运动角度。试验结果表明神经网络预测出的肘关节运动角度与测角仪测出的实际运动角度最大误差小于1度。

The objective of this study is to identify the motion state of human elbow joint based on the EMGs. EMGs were collected from the biceps and triceps muscles of normal subjects when they moved their elbow flexion-extension with time-varying loads. The raw EMG signals were processed and the new defined characteristic was picked up. A four-layer feed-forward neural network model with the characteristic as its input was developed; the weighted values of the model were optimized with the adjusted back-propagation algorithm. By training the model can map the transformation： from the processed EMG signals to the elbow joint angles. The experimental results showed that the maximal error between the joint angle predicted by the network and the actual joint angle measured by the goniometer is less than 1 degree.