sEMG (surface electromyography) signals are physiological signal closely related to human movement, and the analysis of sEMG signals play an important role in the field of human-machine interaction. Aiming at the difficulty of both efficiency and accuracy of electromyographic signal classification, an upper limb sEMG classification method was innovatively proposed, which combined feature screening with classifier hyperparameter optimization. BPSO (binary particle swarm optimization) algorithm was adopted to screen the features. PSO (particle swarm optimization) algorithm was further utilized to adjust the hyperparameters of the LSSVM(least-squares support vector machine). By collecting sEMG signals from four parts of the human upper body and extracting 48-dimensional features from them, classification experiments were conducted on four common movements of upper limb. The results show that the BPSO-PSO-LSSVM algorithm retains only the 21-dimensional features of the EMG data, and the average classification accuracy obtained reaches 97.54%. It is proved that this method can effectively screen out the optimal combination of features for upper limb motion classification and improve the accuracy of movement classification.