Prosthetic hands are becoming more and more popular among people without a hand and it is able to perform certain manual operations in function. In biology, action potentials trigger muscle contractions, and surface electromyography (sEMG) signals are the sum of action potentials under the skin exposed by the electrodes (via sebum and solution conduction). Myoelectric control is a technique that relies on extracting information contained in SEMG signals to determine muscle contractions and then control peripheral devices [1][2][3]. However, how to collect sEMG signals effectively is the first difficult problem for scholars [4] because of weakness, randomness and low signal-to-noise ratio of the sEMG signals. Multi DOF prosthetic hands have high flexibility and then more control parameters are needed. Therefore, the key techniques of EMG control of anthropomorphic prosthetic hand are extracting appropriate control parameters and studying real-time and reliable control algorithms.At present, single-degree-of-freedom prosthetic hands are one of the mainstream of commercial prosthesis markets. With mature EMG signal processing and control system, the performance of them is stable, but the flexibility is not enough. There are also some multi DOF prosthetic hands † These authors contributed equally to this work. * Corresponding authors (email: gaow13975@163.com; shiyan@buaa.edu.cn) that have been commercialized, such as the I-limb prosthesis introduced by Touch Bionics (Scotland) with 5 independent fingers, 24 degrees of freedom. But it's so expensive that ordinary people can't afford it. researchers have proposed a variety of myoelectric control schemes, including threshold based decision, amplitude-based coding method, hierarchical control decision, etc. [5,6]. The thresholding method is that an EMG signal caused by muscle contraction is firstly corrected, filtered and modulated, that then its peak value is compared with the threshold(s), and that finally two or three states of output correspond to the prosthetic hand to relax, bending or stretching, respectively. The scheme is so simple and reliable that it is the control method for a single-degree-of-freedom prosthetic hand. However, due to less information given by each electrode, it is not suitable for the control of a multi DOF prosthetic hand.Like the thresholding decision, in the amplitude-based coding scheme [7], the amplitudes of sEMG collected by differential electrodes are divided into three states, and the artificial hand motion commands are determined by encoding outputs according to time sequence. This decision is simple, whereas requires a lot of training, with low real-time and inflexible control. About the hierarchical control decision, the hand grasp modes of muscle stump are recognized by using advanced signal modulation and pattern recognition al-