Abstract: It appears a big challenge when the multi-DOFs prosthetic hand is controlled by the electromyography (EMG) signals. A novel recognition method of the hand grasp preshaping patterns is proposed to the HIT-DLR prosthetic hand’s EMG control. A new online detection method is designed to collect the accurate onset EMG signals of the grasp preshaping, which uses the Teager-Kaiser energy (TKE) operator and post processing to enlarge the changes of the EMG signal and deal with the spike noise, respectively. Focusing on 4 types of the hand preshaping patterns, different data segmentation methods, different features coming from the time-domain, frequency domain and time-frequency domain, and various classifiers are attempted to find the best classification accuracy. The waveform length and support vector machine are chosen, which can reach an accuracy of 95% and a response time less than 300 ms. The experiment of the prosthetic hand control shows that the hand can swiftly grasp the objects with various shapes.

Key words: Electromyography control, Grasp preshaping, Pattern recognition, Prosthetic hand