Development of an SEMG-Handgrip Force Model Based on Cross Model Selection

摘要

In this paper, a crass model selection (CMS) technology is proposed to accurately estimate the handgrip force from the surface electromyography (SEMG). The proposed model is based on a crass model and a sequence combination analysis (SCA). The cross model is built by sampling the data collected from constant force training tasks (isometric muscle contraction) and force-varying training tasks (gradual dynamic muscle contraction). Isometric contraction data are used to build SEMG-force relationships with regularization. Then, the cross-validation (CV) technique is employed by using gradual dynamic contraction data to obtain the optimal regularization parameter. According to the zero-approaching characteristics of the parameter selected by CV, this paper proposes an SCA method to select a suitable combination of coefficient terms for the CMS model. Through the constant force and force-varying validation tasks, the experimental results showed that the handgrip force estimated from SEMG using the proposed model has higher accuracy than the former models.