Upper Limb Motion Intention Recognition Based on sEMG and EMD-LSSVM

Abstract A method is proposed to enhance the precision of recognizing upper limb motion intentions using surface electromyography. This approach combines empirical mode decomposition with a least squares support vector machine. This method performs empirical mode decomposition on the preprocessed sEMG to obtain its intrinsic mode function and residual function, this method assesses the correlation between each intrinsic mode function and sEMG signals, and selects the component with high correlation as the effective component. Finally, the time-domain features of sEMG and the first three intrinsic mode functions are calculated and combined to obtain a new feature vector. The findings indicate that using EMD-LSSVM improves the accuracy of recognizing upper limb motion intentions by 9 % and 5 % higher than that of EMD-SVM and EMD-PSO-SVM respectively.