Hybrid sEMG Classifier Methods of Lower Limb during LowLevel Flexion

Abstract

Demographic change is becoming an identical serious problem in the world. To improve Quality of Life (QOL) of the lower limb disabled and elderly people, robotic and biomedical researchers have been trying to combine their techniques into the enhancement of rehabilitation system. Robotic devices and sensorimotor functions are important to the physical rehabilitation during the retrieval process. Their requirement is quickly retrieval of lower limb disability state again and improve their living condition very effectively manner. Consequently, human bio-signal, such as surface Electromyogram (sEMG) play a major role to develop assistive devices to improve their recovery process. The aim of this research was identified four movements, such as heel up, toe up, left side up, and right side up of lower limb during low-level flexion. Gastrocnemius Medialis (GM), Tibialis Anterior (TA), Soleus (SL), Flexor Digitorum Longus (FDL) and surface electrode channel 1-4 were assigned correspondingly into each appointed lower limb muscles. Mean Absolute Value (MAV), Variance (VAR), Standard Deviation (STD), Root Mean Square Value (RMS) and maximum amplitude length (MAX) were expended to process and analyze the feature extraction of time domain. The Support Vector Machine (SVM) was used as a first classification method and Linear kernel function, polynomial kernel function, radial basis function (RBF), and sigmoid kernel functions were used to analyze the first part of experiment. RBF was performed higher classification accuracy and it was selected to the pattern classification. Hidden Markov Model (HMM) was followed to handle the second classification task correspondingly. The final classification task was performed to analyze the signal data using hybrid classifier method of SVM with HMM. The pattern recognition accuracy of SVM with RBF indicated good percentage accuracy during the first part. Hybrid classifier has also indicated good percentage level as a second part. When comparison of the both classification methods, the accuracy of hybrid classifier method was displayed higher percentage level than method of SVM classifier. Therefore, according to the resulted data it can be concluded that the hybrid classifier method could be used to obtain higher accurate pattern recognition than other single classification methods. The research was beneficial to the EMG signal researchers to study of hybrid EMG classification technique.