Surface electromyography(sEMG) signal directly and objectively reflects the functional status of nerve and muscle, which has been widely used. In this paper, a sEMG acquisition circuit is designed and used as single channel circuit to collect sEMG signals of five kinds of upper limb movements, then six kinds of features (one of which is quoted from the feature extraction method based on wavelet transform) are extracted by wavelet packet transform(WPT) combining with KPCA, and finally recognition is performed with BP neural network and SVM. Feature extraction based on wavelet transform is also performed for comparison and the difference between PCA and KPCA on feature transform is also studied. The results show that among the six kinds of features extracted by wavelet packet transform, five kinds of recognition rates exceed 95.7%, and the average recognition rate of the high-low frequency coefficients combination feature quoted is more than 99% with a BP neural network. Overall, the recognition rates are high. And the five kinds of features extracted by wavelet transform combining KPCA also achieve a decent recognition rate. The results prove that the sEMG signals collected and the feature extraction method used in this paper are both effective.