EEG Analysis Based on Independent Component Analysis and Positive Time-Frequency Distribution
昇 中迫, 久 吉田, 伸行 山脇 · Institutional Repositories DataBase (IRDB) · 2007
This paper describes a new method for analyzing electroenc~phalogram(EEG) signals combining a noise reduction and a time-frequency analysis.More concretely, a practical scheme for N-dimensional(abbr., N-D) Blind Source Separation (BSS) is first introduced for the purpose of eliminating the effect of artifacts on the observations.The introduced BSS method has two stages: Principal Component Analysis (PCA) and Independent Component Analysis (ICA).The former reduces a dimension N to M (M ::; N) of the observations and ortho-normalizes them, and the latter makes the observations statistically independent based on the evaluation function in terms of the Hermite moments of higher orders.The optimal rotation that transforms M-D ortho-normalized mixed signals into a possible of M independent components is searched for by the gradient method.Then, we apply the proposed method to actual EEG signals.The EEG observations without artifact can be reconstructed considering the inverse of separation process and replacing by zero the independent components regarded as an undesired noise.Finally, the reconstructed (noise-removed) EEG signals are analyzed by the positive time-frequency distribution.We discuss the noise reduction performance of our method and provide a brief comment on our future work.