Comparative analysis of characteristic wavelength extraction methods for nondestructive detection of microplastics in wheat using FT-NIR spectroscopy

Microplastic detection has been acknowledged as challenging so far. Despite advancements in rapid detection methods for analyzing environmental microplastics, limited research has been conducted on detecting microplastics in food substrates. The objective of this study was to investigate the feasibility of utilizing Fourier near-infrared (FT-NIR) spectroscopy optimized characteristic model for quantitative detection of polystyrene (PS) microplastics in flour. A Fourier transform infrared spectrometer was employed to gather spectral information on flour with varying concentrations of PS. Four variable selection methods, namely iterative variable subset optimization (IVSO), bootstrapping soft shrinkage (BOSS), Interval variable iterative space shrinkage approach (IVISSA), and variable-dimensional particle swarm optimization movement window (VDPSO-CMW), were introduced to select features from the preprocessed near-infrared spectrum. Detection models based on partial least squares (PLS) were constructed with the aim of achieving quantitative detection of PS in flour, and comparisons were conducted to evaluate the detection performance of the four models. The VDPSO-CMW-PLS model demonstrates the highest level of generalization performance, according to the research findings. The coefficient of determination ($R_p^2$) is 0.9810, the root mean square error of prediction (RMSEP) is 0.0462%, and the relative percent deviation (RPD) is 7.3890. The research findings indicate that the constructed PLS detection model, utilizing FT-NIR spectral optimization characteristics, can rapidly and accurately detect PS in flour. This study presents a novel technical approach for the prompt quantitative identification of microplastics in food.