A Robust Extraction Algorithm Based on ICA Neural Network

Abstract

Independent component analysis (ICA), blind source separation (BSS) and related methods like blind source extraction (BSE) are all the promising unsupervised neural network technique for analysis of biomedical signals, especially for ECG, EEG and fMRI data. However, most of source extraction algorithms based on ICA neural network are not suitable to extract the desired signal since these algorithms are not to obtain the desired signal as the first output signal. In this paper, we propose an algorithm based on ICA neural network that can extract a desired source signal as the first output signal with a given kurtosis range. Because of adopting a robust objective function, the algorithm becomes very robust to outliers and spiky noise. Simulations on artificially generated data and real-world ECG data have shown that the algorithm can achieve satisfying results.