An Outlier Determination Method for Near-Infrared Spectroscopy Based on the Simplified Orthogonal Distance

作者：Meng Dan-rui; Fu Bo; Xu Ke-xin; Liu Rong^*

来源：Spectroscopy and Spectral Analysis, 2018, 38(4): 1053-1058.

DOI：10.3964/j.issn.1000-0593(2018)04-1053-06

• Near infrared
• Outlier
• Orthogonal distance
• Robust principal component
• Collapse value

Fast detecting and eliminating the outliers is of great significance to improve the reliability of the near-infrared(NIR) spectroscopy analysis. In this paper, the principle of outlier determination method based on orthogonal distance and robust principal component analysis was introduced firstly with the analysis of its limitations. Then an outlier determination method based on the simplified orthogonal distance was proposed, where the spectra of the samples with high concentration were employed to estimate the first robust principal component directly and the statistical parameters of the orthogonal distance were obtained with repeated measurements to detect outliers. Finally, the outliers caused by the temperature fluctuations in the NIR transmission spectra of glucose aqueous solutions and 2% Intralipid solutions, were determined by these two methods. Results showed that, for the orthogonal distance combined with robust principal component analysis method, all the outliers induced by temperature variations could be correctly determined under the collapse value of 40%, while the false negative rates for the glucose aqueous solutions and Intralipid solutions under the collapse value of 25% were 54. 5% and 72. 7%, respectively. Besides, all the outliers induced by temperature variations also could be recognized with the method based on the simplified orthogonal distance, which saves the need for collapse value and shortens the tine for measurement. Therefore, the outlier determination method based on the simplified orthogonal distance is more practical than the robust principal component analysis.

• 出版日期2018-4
• 单位天津大学