sEMG Pattern Recognition of Muscle Force of Upper Arm for Intelligent Bionic Limb Control

Two new feature extraction methods, window sample entropy and window kurtosis, were proposed, which mainly aims to complete the surface Electromyography (sEMG)-muscle force pattern recognition for intelligent bionic limb. The inspiration is drawn from physiological process of muscle force generation. Five hand movement tasks were implemented for sEMG-muscle force data record. With two classical features: Integrated Electromyography (IEMG) and Root Mean Square (RMS), two new features were fed into the wavelet neural network to predict the muscle force. To solve the issues that amputates' residual limb couldn't provide full train data for pattern recognition, it is proposed that force was predicted by neural network which is trained by contralateral data in this paper. The feasibility of the proposed features extraction methods was demonstrated by both ipsilateral and contralateral experimental results. The ipsilateral experimental results give very promising pattern classification accuracy with normalized mean square 0.58 +/- 0.05. In addition, unilateral transradial amputees will benefit from the proposed method in the contralateral experiment, which probably helps them to train the intelligent bionic limb by their own sEMG.