Enhancing Honey Adulteration Detection With Optimal Subspace Wavelength Reduction in Vis-NIR Reflection Spectroscopy

Abstract

The accurate detection of honey adulteration is paramount for maintaining the quality and authenticity of honey products. In this study, we introduce a novel feature selection method, termed Optimal Subspace Wavelength Reduction (OSWR), and integrate it with reflectance Visible-Near Infrared (Vis-NIR) spectroscopy to enhance the discrimination between pure and adulterated honey and predict adulteration levels. OSWR efficiently addresses the dimensionality challenge of large spectral datasets, reducing 2151 wavelengths to a compact and informative set of 39 wavelengths. We comprehensively evaluate machine learning (ML) models, focusing on OSWR as a pivotal component of our methodology. Our results reveal remarkable success in discriminating among pure honey, adulterated honey, and sugar syrup, with an impressive classification accuracy of 96.67% achieved using OSWR, coupled with Standard Normal Variate (SNV) preprocessing, Linear Discriminant Analysis (LDA) feature extraction, and K-Nearest Neighbors (KNN) classification. Furthermore, this study demonstrates the effectiveness of OSWR for predicting adulteration levels, where it achieves an accuracy of as high as 92.67% when coupled with SNV, LDA, and KNN. This work highlights the potential of OSWR as a feature selection method in the context of honey adulteration detection. Through the integration of Vis-NIR spectroscopy and OSWR, our approach offers a tool for enhancing honey products' quality and authenticity assessment, potentially simplifying spectral data analysis.