基于肌电信号的人手姿态多模式识别方法

Study on Recognition of Multi-mode Hand Gestures Based on Myoelectric Signal

为了实现多自由度假手的控制,需要基于人体肌电信号识别更多的手部姿态.采用6枚表面肌肤电极获取肌电信号,使用样本均值构成特征向量训练支撑向量机,通过对人手姿态模式进行合理规划,实现了人手19种姿态的有效分类.相比传统稳态特征集而言,新方法使用阈值特征集训练分类器,使其在总体及模式过渡特征的识别率上均有提高.基于此而构建的人手姿态多模式在线识别方法将使多自由假手的肌电控制更加直观与有效.

The development of multi-DOF prosthetic hand appeals for classfying more hand gestures based on myoelectric signals extracted from the forearm. Six surface electromyography （EMG） electrodes were used to acquire myoelectric signals. Each channel＇s sample means are used to constitute feature veztors for training the support vector machines （SVM）. Then 19 modes of hand gestures can be discriminated effectively. Comparing with some traditional training methods that use the steady-state features of myoelectric signals, the new method improves the predicting accuracy both on full scale features and the ones between the mode transitions. This will benefit the on-line recognition of the hand gestures, therefore make the multi-DOF prosthetic hand＇s EMG control more intuitive and effective.