Time series modeling of surface EMG based hand manipulation identification via expectation maximization algorithm

In this paper, we focus on the method of employing the expectation maximization (EM) algorithm to the modeling of surface electromyography (sEMG) signals based on hand manipulations via available time series of the measured data. The model for the sEMG is developed as a hidden Markov model (HMM) framework. In order to represent dynamical characteristics of sEMG when multichannel observation sequence are given, a stochastic dynamic process is included in it based on the maximum likelihood estimation (MLE) principle. By using the EM algorithm, the hidden model parameters and the feature of the signal can be identified easily. Ten people of different time series data sets of different hand grasps and in-hand manipulations captured from different subjects are collected. The two different classifiers were used to recognize these hand manipulation signal. Compared with time and time–frequency domains and their combination feature, the proposed algorithm of the inferred model gains better performance and demonstrates the effectiveness. The average identification accuracy rate is 93% and the maximum classification ratio is 100%.