Lower Limb Activity Prediction Using EMG Signals and Broad Learning

Abstract

This article provides an improved lower limb activity prediction system using surface EMG raw data and Broad Learning (BL) classifier. The proposed feature is calculated using three main sequential steps; First, convert EMG raw data to several narrow overlapping segments; Second, apply Kaiser window function and short-time Fourier transform for each segment; Third, find the texture analysis of EMG power spectrum. The public UCI database is used for system evaluation. Experiments show that lower limb activity prediction achieved the highest results of 96% 94%, and 90% for knee abnormal group, normal group, and both groups together, respectively. Moreover, This study proves the possibility of achieving an acceptable activity prediction results in case of mixing normal and knee abnormal groups together.