Wavelength Selection of Terahertz Time-Domain Spectroscopy Based on a Partial Least Squares Model for Quantitative Analysis
Qingxiao Ma,Chun Li,Biao Wang,Xin Ma,Ling Jiang
DOI: https://doi.org/10.1364/ao.427238
IF: 1.9
2021-01-01
Applied Optics
Abstract:Terahertz spectroscopy, combined with chemometric methods, has proved to be an effective tool in the quantitative analysis of many substances. However, current research has mainly focused on comparing different algorithms under the full spectrum. In fact, the full spectrum is not only composed of characteristic features of the samples, but also many types of noises. Hence, the accuracy of the quantitative analysis may be unsatisfactory if the full spectrum is selected. In this paper, a wavelength selection method based on partial least squares and absorption peak was proposed and an efficient frequency band was determined in the quantitative analysis of three types of pesticides, i.e., 6-benzylaminopurine, 2, 6-dichlorobenzonitrile, and imidacloprid. By introducing two parameters, the sum of peak intervals ( S i ) and peak width, the most efficient peak was selected from multiple peaks and the specific peak width was given with the aid of particle swarm optimization. We concluded that the most efficient absorption peak for quantitative analysis corresponding to the largest S i and full width near one-half maximum could characterize full spectrum information precisely. Comparing before and after wavelength selection, the correlation coefficient ( R ) of the three pesticides have increased from 0.9671, 0.9705, 0.9884 to 0.9921, 0.9934, and 0.9957. In conclusion, the proposed wavelength selection method was demonstrated to be very efficient for the quantitative analysis of the pesticide mixtures, which also could be helpful in the analysis of other multicomponent mixtures with absorption peaks.