Control method for myoelectric hand using convolutional neural network to simplify learning of EMG signals

By employing learning, myoelectric hands can demonstrate high performance, but learning takes a long time. This problem reduces the motivation of a myoelectric hand wearer and introduces an adverse effect on the mastery of myoelectric hand operation. Thus, we propose a control method for myoelectric hands that uses convolutional neural network without relearning when a wearer reseat the myoelectric hand. As a result of the experiments using a healthy subject, the proposed method was shown to maintain a better classification rate than the traditional method even if the methods learn using datasets measured in other days. This advantage will reduce the relearning process of myoelectric hands when a wearer reseat the hand.