Due to the limited accuracy of time-domain and frequency-domain features of surface electromyography (sEMG) signals in gesture classification, a novel method was proposed based on variational mode decomposition (VMD) for sEMG signal decomposition, combined with a particle swarm optimization-support vector machine (PSO-SVM) algorithm for gesture classification. First, the sEMG signals were preprocessed and decomposed into multiple scales using VMD. Features such as approximate entropy, energy entropy, permutation entropy, sample entropy, singular spectrum entropy, and spectral entropy were extracted. Next, dimensionality reduction was performed on the extracted features using correlation analysis, random forest, and Lasso methods to eliminate redundant information. Finally, the reduced feature set was classified using the PSO-SVM algorithm to achieve efficient and accurate gesture recognition. Experimental results demonstrate that the proposed method achieves a recognition accuracy of 97.30% for six gesture types, showcasing its high classification performance and strong potential for practical applications.