Fast Quantitative Analysis of Hidden Dangerous Substances in Mail Based on Specific Interval PLS

• Published in: Journal of infrared, millimeter and terahertz waves Vol. 42; no. 5; pp. 572 - 587
• Main Authors: Li, Tao, He, Jian-an, Zhang, Liang, Ye, Ying, Gu, Dayong, Zhang, Sixiang, Zhang, Pengjun, Hu, Xuenan, Wei, Shuang
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.05.2021; Springer Nature B.V
• Subjects: Classical Electrodynamics; Electrical Engineering; Electronics and Microelectronics; Engineering; Inspection; Instrumentation; Least squares; Mail; Model accuracy; Multiple regression models; Multivariate analysis; Principal components analysis; Qualitative analysis; Quantitative analysis; Root-mean-square errors; Spectra; Mail; Terahertz; Time-domain spectroscopy; Detection; Quantitative analysis
• ISSN: 1866-6892; 1866-6906
• DOI: 10.1007/s10762-021-00790-x

Terahertz radiation has many unique characteristics that make it useful for noninvasive mail inspection. While qualitative analysis of mail for suspicious objects is a relatively instantaneous process, quantitative analysis methods may be time-consuming. Multivariate analysis methods, including principal component analysis (PCA), partial least squares (PLS), and interval partial least squares (iPLS), were used for quantitative model building and to predict the content of substances in mail. The optimal spectral interval was selected by analyzing the influence of different spectral regions on the predicted results. A specific interval partial least squares (SiPLS) model was established to improve prediction accuracy and reduce the root mean square error (RMSE) by an order of magnitude. The content of dangerous substances was calculated using SiPLS, established by referencing spectral data of pure substances. Our methods demonstrated that establishing a multiple regression model based on spectral data of pure substances could predict the content of dangerous substances in mail.