Prediction of Lower Limb Action Intention Based on Surface EMG Signal

Abstract

Aiming at the problem of helping lower limb disabled people recover their ability, this paper classifies and recognizes four kinds of movements (sitting, thigh lifting, leg lifting and leg straightening) based on surface electromyography (sEMG). Firstly, the healthy subjects were trained by collecting the surface EMG signal data of multiple groups of lower limbs. The EMG sensors were placed in the rectus femoris, medial femoris, semitendinosus and gastrocnemius to collect data. The noise is filtered by Butterworth filter and EMD signal reconstruction method, and the pure signal is obtained. The EMG features of each channel data are extracted, and the normalized input vector is sent to the classifier. In this paper, traditional classifiers such as random forest, xgboost and linear discriminant analysis are used to classify lower limb movements. Then, the accuracy of various classifiers is compared. It is found that the recognition accuracy of machine learning is higher, and EMD signal reconstruction method is better than Butterworth filter in the pretreatment of EMG signals, LDA classification accuracy is the highest, which can reach 100%. At the same time, the prediction speed of machine learning is faster, which can reach 300ms.