Position estimation for elbow joint by classification of surface EMG signals with deep learning methods

Classification of surface electromyography (sEMG) signals, which contain movement information of muscles, are used to achieve a natural working order in systems for rehabilitation. In this thesis, sEMG signals of upper extremity biceps and triceps muscles were classified with Long Short Term Memory (LSTM) artificial neural network which is a Deep Learning method, in order to estimate elbow joint angle. LSTM model successfully and effectively made elbow joint angle estimation by obtaining high accuracy using varying average value and envelope peak value of sEMG. The results are compared with the elbow joint angle estimation of the Multi-Layer Sensor (MLP) which uses the attributes of the sEMG signals. The designed LSTM model has shown that it can estimate elbow joint angle more effectively in terms of calculation, by obtaining accuracy as high as MLP.