小波包分解结合异常值检测自动去除眼电中眨眼干扰的方法
A method for removing blink signals in electrooculogram by using wavelet packet decomposition and outlier detection

为减少有用眼电信号的损失，提出一种小波包分解和异常值检测(WPT-OD)去除眨眼信号的新方法. 该算法首先利用小波包方法将原始信号进行分解，得到低频分量进行重构；然后应用异常值检测中三种常用的准则，即肖维勒准则、 皮尔斯准则和修正箱线图法确定眨眼信号的区域，并将该区域置零. 实验发现，WPT-OD的平均正确率达到98.9％，其中修正箱线图法效果最好，其去眨眼信号与原始信号相关性高达95.33%，损失率仅为4.17%. 实验表明：WPT-OD算法能够准确地确定无意识眨眼的起点和终点，可保留更多的有用信号且与原始信号的相关性强.
In order to reduce the loss of useful eye signals，this paper presents a new method of wavelet packet transform and outliers detection(WPT-OD) to remove blink signals. Firstly，the wavelet packet method is used to decompose the high frequency components from the original signal，and then the three commonly used criteria of the anomaly detection are applied. The position of the blink signal was determined by Chauvenet criterion(CC)，Peirce criterion(PC) and ADJBP(adjusted box plot)，and it was set to zero. It was found that the average correct rate of WPT-OD was 98.9%. The correlation between the blink signal and the original was up to 95.33% and the loss rate was only 4.17%. The experimental results verify the effectiveness of the algorithm. The WPT-OD algorithm can accurately determine the starting and ending points of involuntary blinks，and it can retain more useful signals and have a strong correlation with the original signals