A Pattern Recognition Method for Surface Electromyography Based on Nonlinear Features

A new method named ball-averaged Lyapunov exponents method is presented to calculate Lyapunov exponents of nonlinear time-series signals. The method can be used as feature extraction and classification of electromyography. Firstly, the Lyapunov exponents of electromyography is calculated and it is combined with correlation dimension as input eigenvector. Then, multi-class classifier is constructed based on Twin Support Vector Machines (TSVM) with binary-tree. Finally, the four hand gestures (namely, radial flexion and ulnar flexion, hand opening and closing) are classified. The experimental results show that the method has stronger anti-jamming capability than Rosenstein method, and the recognition rate is above 96.0% in feature extraction and classification of electromyography. The proposed method is suitable for analyzing chaotic signals with lower signal-to-noise ratio.