표면 근전도 신호 분석을 위한 필터와 분류 방법의 최적 조합에 대한 연구

Abstract

Action potential occurs when the muscles contract, which is propagated along the muscle fiber, called EMG. EMG is also transmitted along the skin layer and measured, which is called sEMG. Since the sEMG has random characteristics, appropriate selection of filters and classifiers is required to classify them into physical motion. Hence, we would like to compare classification accuracy between the true and predicted values of the classification to find out which combination has the best performance. Among various filter methods, MAV, RMS and Bayesian filtering were used and representative methods for machine learning and deep learning, i.e., SVM and LSTM, respectively, were selected to be applied. In the experiment, the forearm is divided into four parts and 16 electrodes are attached. Then, we have collected the data during 10 repetitions of the six movements related to hand movement and we conducted 10-fold cross validation.