In recent years, human-computer interaction technology has become increasingly mature, and equipment based on human-computer interaction technology, such as mechanical exoskeletons, has also begun to be used in military, medical and sports rehabilitation fields. Among them, myoelectric signal is the cornerstone of controlling human-computer interaction devices, and its feature extraction and classification recognition quality are related to the performance of such devices. Nowadays, the EMG signal processing module still has shortcomings in signal preprocessing, feature extraction and classification recognition, in view of this problem, this thesis designs a surface electromyography signal (sEMG) classification recognition system, and innovatively improves the Ensemble Empirical Mode Decomposition method(EEMD) in the signal processing process. A two-layer ensemble classifier using the support vector machines (SVM) algorithm and the gradient boosting decision tree (GBDT) algorithm built by stacking fusion strategy is designed to improve the accuracy of human motion classification recognition based on myoelectric signals.