A Low-rank strategy for improving the prediction accuracy of partial least square models

• DOI: 10.1016/j.infrared.2021.103798
• Publication Date: 2021-06-04T23:06:24Z
• Authors: Qifeng Li; Yuanlin Dai; Jinglai Sun; Yangguang Han; Xiaoran Fu; Yunpeng Yang; Xiangyun Ma; Huijie Wang
• Keywords: 01 natural sciences; 0104 chemical sciences

Abstract Infrared spectroscopy has been widely used in fast and non-destructive quantitative analysis fields. However, some physical phenomena in the infrared spectra will limit the prediction accuracy of the quantitative analysis. Due to the high correlations of the spectral signatures, the infrared spectral dataset has the low-rank property, which can be used as a constraint to remove the undesired variations of the infrared spectra. In this paper, a low-rank strategy for improving the prediction accuracy of Partial Least Squares (PLS) chemometric model is proposed. The low-rank PLS (LR-PLS) method is used for the quantitative analysis based on the infrared spectra of different samples. Compared with the traditional methods, the proposed method has better performance in improving the prediction accuracy.