Study On Independent Component Analysis And Its Application In Event-related Potential

Electroencephalograph (EEG) is a kind of spontaneous or induced electrical activities recorded from the electrical potentials generated by nerve cells in the cerebral cortex. As a special kind of invoked EEG--ERP reflects the cerebral nerve electrical activity during the cognitive process; it is a powerful tool that being used to do research in advanced human cognitive functions. However, the ERP signals are very weak and often submerged in the spontaneous EEG, meanwhile it's also interfered by many other noises easily. For instance, EOG, EMG, ECG and other all kinds of rhythm waves. So for a long time, how to extract useful ERP signals from high noise background is the key to ERP signal processing. ICA is a kind of newly developed multidimensional signal processing method. Its research object is independent gauss signals. Under certain conditions, ICA could extract implicit independent from multi-channel signals that measured synchronously.This paper firstly introduces the development process and study methods of EEGThen it introduces the concept of ERP, classic components, experiment mode, method and extraction of ERP. In the second half part of the paper, I studied the basic theory and algorithm of ICA, and apply the FastICA algorithm to the experiment of ERP denoising and extraction of ERP (P300 mainly) successfully. And do the following several particular aspects:(1)Connects 60 periods of induced EEG data, and decompose it through FastICA algorithm, Extract noise components with prior knowledge of EEG. So the recorded EEG signals are denoised. (2)Based on the FastICA algorithm before, mainly studies and discuss about next two methods.One method is extracting ERP composition using average superposition method based on the FastICA algorithm. Another method is to extract ERP composition using less time average superposition.The experiment data comes from the institute of psychology and behavior of Tianjin normal university. I have done the experiments with both single data and group data respectively. Then compare two methods with the traditional average superposition. It proves that FastICA algorithm can effectively extract ERP, also can extract ERP rapidly. Finally the experiments prove that both of the methods can achieve good results in extracting ERPs.