Classification of Muscular Strength during Palmar Grasp Exercises Using Surface EMG Signals

Several studies have been carried out for strength detection, using muscle models or machine learning algorithms with surface electromyography (sEMG) signals. However, some limitations have been encountered such as the need of measurements of joint angles, among others. This paper presents a method, based on a sEMG signal processing algorithm and a machine learning algorithm, which requires no additional sensors to classify muscular strength during palmar grasp exercises. An experimental protocol with 7-healthy-subjects was conducted in order to acquire sEMG signals during this type of exercises, with four levels of strength. Subsequently, an offline sEMG signal processing algorithm extracts 21 features in the time-domain, frequency-domain, and time-frequency domain. Subsequently, a dimensionality reduction algorithm and a machine learning algorithm were implemented. The best inter-subject test obtained a mean classification correct rate (CCR) of 0. 68 0.04. The intrasubject tests obtained a mean CCR of 0. 71 0.04. As a result of this study, it is possible to propose a method for the classification of muscular strength during palmar grasp exercises using sEMG signals. However, some important factors, such as temperature and others, were not controlled during the trials. Therefore, it is considered that it could be one of the possible causes that the CCR was not greater than 71%.