EEG ENHANCEMENT USING EXTENDED KALMAN FILTER TO TRAIN MULTI-LAYER PERCEPTRON

摘要

In many applications of signal processing, especially in biomedicine, electroencephalogram (EEG) is the recording of electrophysiological brain activity along the scalp over a small interval of time and it is a biological non stationary signal which contains important information. Analysis of EEG signal is useful to identify physiological situations of the human as normal and epileptic subject. EEG signal becomes more complicated to be analyzed by the introduction of the noise. In this paper, a nonlinear Kalman Filter scheme where an extended Kalman filter (EKF) based Multi-layer perceptron (MLP) model is proposed to remove white and colored Gaussian noises from EEG recordings in physiological and pathological states (normal and epileptic). The MLP is one of the artificial neural network (ANN) models that has great track of impacts at solving a variety of problems. Activation function is one of the elements in MLP neural network. Selection of the activation function as sigmoid in the MLP network plays an essential role on the network performance. Thus, the MLP parameters as weights, and outputs are trained by an EKF in order to minimize the difference between the output of the neural network and the desired outputs. The results comparison studies are evaluated with root mean square difference (RMSD) and signal to noise ratio (SNR). The elapsed time is decreased using this method compared to normalised least mean square (NLMS) and Meyer wavelet methods. These parameters applied to EEG signals show the validity and effectiveness of the proposed approach.