Quality Control and Low-Quality Detection in Packaging using Near-Infrared Spectroscopy

Abstract

Using Near-Infrared Spectroscopy as a method of quality determination and to identify low-quality packaging in the food sector. An analysis is conducted using a portable NIR spectroscopy tool (NIRvaScan) on 244 NIR wavelengths collected from high- and low-quality packaging materials in the spectrum between 900 - 1700nm. These wavelengths are analyzed for quality distinguishing features using Linear Discriminant Analysis, Principal Component Analysis, and Partial Least Squares. The following classification algorithms are discussed: DecisionTree, XGBoost RandomForest, and Support Vector Machines. A thorough analysis of transformation and pre-treatment methods are provided, including the transformation from the Reflectance to the Absorbance scale, and the pre-filtering methods Standard Normal Variate (SNV), Multiplicative Scattering Correction (MSC), and the Savitzky-Golay (SG) filtering algorithm.