Prediction of EMG Signal on Missing Channel from Signal Captured from Other Related Channels via Deep Neural Network

Abstract

Capturing electromyography (EMG) is always time-consuming and tedious work as it has to located muscle group and attach the electrode with proper skin preparation. In order to reduce the capturing channel, we study modeling of coordinated muscles in this paper. Typical movement are repetitively conducted with seven major muscles on one leg with help of recruited participants. A deep neural network (DNN) is proposed and trained with historical data. The resulting EMG on vastus lateral (VL) are then predicted from other muscles including rectus femoris (RF), semitendinosus (ST) and biceps femoris (BF), tibialis anterior (TA), glutaeus maximus (GM) and soleus (SO). The predicted EMG signals on VL are compared with measuring result and shows the high accuracy. The predicted result has compared with other learning-based method to show its effectiveness. This result can be used for less channel EMG capturing with predicting signals from the missing channel which will save experimental time and money investing on the capturing hardware.