Análise de características do sinal EMG para auxílio em processos de classificação de padrões

Abstract

Electromyography is a technique that consists of the development, capture and study of electromyographic (EMG) signals from muscle activity. Within electromyography, a very important step refers to the features extraction, where useful signal information is synthesized, helping to remove undesired parts and interferences of the same. The present work presents a study containing the analysis of EMG signal features to aid in standards classification processes. For this, several steps were required involving acquisition, preprocessing, features extraction and selection, classification, obtaining and evaluation of results, using the software LabVIEWTM and MATLABTM. The acquisition was performed using an eight-channel armband attached to the forearm. Six gestures were performed (flexion, extension, flexion to the left, extension to the right, supination and pronation). Subsequently the collected signal was submitted to pre-processing, which involved the conditioning and segmentation stages. Features extraction was performed for 26 features of time domain and frequency domain. The features selection involved the use of simple combination for the assembly of groups with 2, 3, 4 and 5 features, in addition to the analysis of characteristics individually. The classification of the six gestures was made using the LDA (Linear Discriminant Analysis) and QDA (Quadratic Discriminant Analysis) classifiers. The results evaluated were related to the accuracy rates of the classifiers (through bar graphs) and the features intensity for the armband channels in relation to the gestures performed (through polar graphs). Finally, through the results obtained it was concluded that the present work made possible a broader analysis of the EMG signal, as well as providing an evaluation of the contribution of different features to the improvement in the performance of movement classification.