即時自動化雜訊去除演算法應用在腦電波分析

Abstract

在腦電波（electroencephalogram, EEG）測量中，由於外在因素的影響，電訊號容易被諸如眼球活動、肌肉活動或環境干擾等雜訊汙染。雜訊不僅會造成腦電波在基礎研究中解釋和分析上的問題，也會造成腦機介面（Brain Computer Interface, BCI）的不正常運作。在過去的基礎研究中，雜訊根據研究者的主觀經驗，以手動的方式去除，這是一項非常費力且耗時的工作。近年來許多研究提出了各種去雜訊的方法，然而，這些方法大多無法即時去雜訊、或只能去除少數幾種雜訊，導致這些方法不易應用在現實生活中。 本研究提出基於典型相關分析（canonical correlation analysis, CCA）的即時自動化雜訊去除演算法。典型相關分析是一種盲訊號分離（blind source separation, BSS）的方法，將觀察到的訊號分解為具有最大自相關及最小互相關的源訊號。此外，我們提出一套自動化偵測雜訊成分的演算法，利用雜訊特徵擷取方法，並搭配透過支持向量機（support vector machine, SVM）自動辨識雜訊。結果顯示本方法在不影響腦電波的情況下，能有效即時去除雜訊。

Electroencephalogram (EEG) signals are usually and easily contaminated by interferences from several artifacts, such as eye movement, muscle activity, heart rhythm, and environmental noise. Artifacts not only lead to some problems in interpreting and analyzing the EEG fundamental studies, but also cause the malfunction of a brain computer interface (BCI). In fundamental study, the artifacts used to be eliminated manually and empirically, which was a very inconvenient and time-consuming process. Recently, several artifact removal methods were proposed to remove artifacts. However, most of these methods cannot support real-time implement or can just reject few kinds of artifacts. Therefore, these methods are hard to implement and apply to the real applications. This study proposes a novel method for real-time automatic EEG artifact removal based on canonical correlation analysis (CCA). CCA is a blind source separation (BSS) technique that separates the observed signals into source signals which are maximally autocorrelated and mutually uncorrelated. In addition, this study also proposes an automatic artifact detection algorithm, including a novel artifact feature extraction method and the support vector machine (SVM) classifier. As results, the proposed method can effectively reject artifacts from EEG signals and still maintain the phenomena of EEG activity for VEP and SSVEP tasks.