Investigação da relação dos métodos de extração de características e os movimentos das mãos, no processamento de sinais sEMG

Abstract

There are several techniques for measuring electric and magnetic potentials that originate in the human body. Among these techniques is the electromyography technique, also known as EMG. EMG signals are commonly taken from the human body using non-invasive electrodes (i.e., superficially on the skin) that are placed over the muscles of interest. By processing these signals, it is possible to classify them, i.e. to identify the types of movement that the patient is performing. A good classification depends on previous steps, such as pre-processing and feature extraction. This work aimed to investigate which features performed best and worst in classifying different signals. For this, four datasets containing different hand movements performed in everyday life were processed. At the end of the work, through qualitative and quantitative analysis of the implemented features, it was concluded that some temporal features are not interesting for processing, such as TM3, TM4 and TM5 while others, such as logarithmic detector, stood out. The frequency features only proved interesting on a dataset whose data sampling frequency was much higher than the others.