sEMG-Based Knee Joint Angle Prediction Using Independent Component Analysis & CNN-LSTM

Abstract

In recent years, surface electromyography (sEMG)- based neural decoding has shown prospective applications in rehabilitation medicine and smart prosthetics, and sEMG signals have been increasingly used to operate wearable devices. In order to develop an exoskeleton controller that can assist the human body to walk up stairs, we investigated the relationship between joint angle and surface EMG (including the effect of different algorithms on the predicted results) when the human body walks up stairs. Five subjects with normal joints participated in the experiment. In this paper, a new model-CNN-LSTM (Convolutional Neural Network- Long Short-Term Memory) is proposed to predict the angle of the knee joint. To reduce the crosstalk between different sensors, the ICA (Independent Component Analysis) algorithm was used as a data preprocessing method. The method is shown to be efficient by comparing the prediction results of the algorithms. This is the first step towards myoelectric control of an assisted exoskeleton robot using discrete decoding. The results of this study will lead to the development of future neurologically controlled mechanical exoskeletons that will allow people who need assistance to perform more activities.