Comparison of Deep Learning & Adaptive Algorithm Performance for De-Noising EEG

Abstract

Various forms of artifacts can readily contaminate an electroencephalogram recorded using surface electrodes. A comparison of several electroencephalogram (EEG) de-noising methods is shown here. Five distinct forms of noise are reduced using three different strategies, and the results are compared. These three procedures are Recursive Least Squares (RLS) adaptive algorithm, Least Mean Squares (LMS) method, and Fully Connected Neural Network (FCNN). The results are shown using time-domain plots of the real EEG signal, noisy EEG signal, and forecasted EEG signal. For comparing the performance of the three de-noising techniques here relative-root-mean-square-error (RRMSE) and signal-to-noise-ratio were used. Here, exploring the values of the parameters, we find that FCNN predicts a better result than other two algorithms.