Rapid analysis the type of customs paper using Micro-NIR spectrometers and machine learning algorithms.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Accurate paper type identification is vital for customs control.
• Existing methods for paper discrimination are often impractical for field use due to complexity or equipment requirements.

Purpose of the Study:

• To evaluate the efficacy of Micro near-infrared (Micro-NIR) spectroscopy for rapid customs paper classification.
• To compare Micro-NIR performance against Attenuated Total Reflection-Fourier Transform Infrared Spectroscopy (ATR-FTIR).
• To assess various machine learning algorithms for paper type prediction using spectral data.

Main Methods:

• Utilized a Micro-NIR spectrometer to acquire spectral data from paper samples.
• Employed four classification algorithms: K-nearest neighbor (KNN), Soft Independent Modeling of Class Analogy (SIMCA), Partial Least Squares Discriminant Analysis (PLS-DA), and Least Squares-Support Vector Machine (LS-SVM).
• Validated model performance using Monte Carlo sampling for dataset generation and compared results with ATR-FTIR.

Main Results:

• Micro-NIR combined with LS-SVM achieved the highest average accuracy (98.91%) and best model stability (1.06% standard deviation).
• KNN and LS-SVM demonstrated superior performance with Micro-NIR data compared to SIMCA and PLS-DA.
• Micro-NIR showed comparable or better results than ATR-FTIR, particularly for KNN and SIMCA models.

Conclusions:

• Micro-NIR spectroscopy coupled with machine learning algorithms provides an efficient and rapid solution for customs paper classification.
• This approach is a viable and potentially superior alternative to traditional methods like ATR-FTIR in customs environments.
• The study highlights the potential of portable spectroscopic techniques for on-site material identification.