I. Introduction
EEG contains information about brain structure and function, which makes it increasingly important in the diagnosis and treatment of brain diseases and abnormalities [1]. However, EEG signal is susceptible to various artifacts due to its low amplitude [2]. The existence of artifacts in EEG signals will make it difficult for physicians to establish an accurate diagnosis. Among various artifacts, electrooculogram (EOG) artifacts are the most difficult to remove, which are produced by changes in the orientation of the electrical fields of the eyes [3]. EOG artifacts brings great trouble to the processing, analysis and classification of EEG signal,, and affect the clinical application of EEG. Therefore, it is necessary to use appropriate mathematical algorithm to remove EOG artifacts as much as possible when analyzin gEEG.