Independent Component Analysis And Its Application In EP Signal Extraction And Separation

Brain signal reflects the electricity activities of cerebral tissue and brain function status. Brain signals are classified into EEG signal and EP signal. As the first step of research on brain signals, the removal of artifacts in EEG signals and the extraction of EP signals play an important role.ICA is a multi-channel signal processing method derived from Blind Source Separation. It helps to achieve the enhancement and analysis of signals. We have acquired satified results as to multi-channel exaction in visual evoked potential. On the other hand, ICA is gradually applied to removal of artifacts in EEG and has resolved many problems resulted from traditional method in practical use.Genetic algorithm is based on the evolutionarily stable strategy, the most extraordinary advantage is that this algorithm can carry through whole space research and will not get into local extremum, even under the case that fitness fuction is discontinuous, irregular and noise contaminated. This thesis proposes an evoked potential extraction method based on independent component analysis and genetic algorithm. We first take kurtosis of EP signal as datum mark and take reciprocal of the bias degree beween the kurtosis of other sigal and the datum mark as the fitness function. Then we extract the EP signal by searching the maximum of the fitness function through the whole space, and accomplish fast extraction of evoked potential. Since this method incorporate genetic algorith, it has strong astringency, When comparing to other EP signal extraction method it has many advantages such as fast computing, easily realized, not necessarily to introduce into the reference signals and so on.Since the brain signal collected clinically has abundant frenqency component, it is difficult to seprate them from each other completely only by ICA method, usually there are still much noise left in high frequency of the seprated signal. Considering the advantage of removal of noise by wavelet transform, if we do filter the noise from the mixing signal before we do further signal processing by ICA method, the seprated result is improved evidently. That can overcome shortcomings of present methods on noise undistinguishable. This thesis incorporates wavelet transform with ICA method based on genetic algorithm, and applicates them to evoked potentional extraction, we get satisfactory results.