Abstract
This paper describes a human forearm motion discrimination method by the concise neural network for myoelectric hand control. In the conventional studies, the neural network is often used to estimate the motion intention by the myoelectric potential and realize the high discrimination precision. However, the feature extraction processing of the myoelectric potential and the neural network structure tends to become complicated. Therefore, they are not suitable due to the calculation amount increase because the myoelectric hand system is difficult to be equipped with a high-speed CPU. The purpose of this study is to obtain the high discrimination precision without complicating the neural network structure. This study proposes the optimization method of "input data" to Neural Network and "data for learning". Therefore, the total system will be concise and inexpensive. In addition, this study proposes the robust relearning method against the myoelectric potential change by the muscle fatigue. Some experiments on the myoelectric hand simulator show the effectiveness of the proposed motion discrimination method.
