Study On Simultaneous Decoding Of Grasp Patterns And Force From Electromyography Signals And Myoelectric Control For A Prosthetic Hand

Myoelectric control realizes the anthropopathy function of operating a prosthetic hand by upper limb amputees’brain. At present, most of multi-degree-freedom myoelectric prosthesis only offer myoelectric control of grasp patterns. But the disabled further hope they can regulate the magnitude of grasp force while prosthesis implement amputees’voluntary motion patterns. With regard to the situation, this paper will study simultaneous decoding of grasp patterns and force from electromyography (EMG) signals and myoelectric control.However, when users of prosthetic hands apply different grasp force, usually the muscle contraction effort (MCE) of amputees’stump changes. That leads to increasing the non-stationary of EMG and the recognition rate of grasp patterns will go down.To solve the problem, firstly, we studied the EMG pattern recognition without consideration the interference from variational MCE. According to the evaluation index of grasp patterns’discrimination and recognition rate, some outstanding features and classifiers were selected preliminarily. The optimum recognition rate is98.06±2.05%.Then, consider the situation of grasp samples with different MCE. This paper proposed a index to evaluate the robustness of EMG pattern recognition method. Based on the index, on the condition of small training set, a robust pattern recognition method was selected.After settled aforementioned problem, force decoding from sEMG was studied. Based on analyzing the EMG features which represent the MCE, two methods of discrete force decoding were compared, that’s the way of classifying and the way of threshold criterion. Analysis results show the latter is better and only need small training set.Thus, the strategy of simultaneous EMG decoding of grasp patterns and grasp force was proposed——pattern recognition with features of sEMG’s cepstrum coefficients combined with slope sign change (SSC) and classifier of linear discriminant analysis (LDA); force decoding method based on setting threshold criterion to decide the level of sEMG’s mean absolute value (MAV). the accuracy of grasping patterns decoding is87.95±5.74%and accuracy of force decoding is72.91±9.58%.At last, based on the strategy of simultaneous EMG decoding of grasp patterns and grasp force, the experimental platform of simultaneous myoelectric control on grasp patterns and force for prosthetic hands was built. On-line experimental results of myoelectric control indicates the strategy to decode the grasping patterns and force from EMG simultaneously is feasible. On the condition of small training set and changing the MCE of grasp patterns, the accuracy of pattern recognition is approximately80%and the method is robust. But the way of force decoding is not stable that caused by sEMG’s non-stationary. Thus the method of force decoding from sEMG need to be improved in the future.