Abstract
Green tea is consumed worldwide for its health-promoting properties, but it remains vulnerable to microplastic contamination during packaging and processing. Microplastics such as polystyrene (PS) and polyethylene terephthalate (PET) pose potential risks to human health and food safety, underscoring the need for effective detection methods. In this study, Surface-Enhanced Raman Scattering (SERS) using gold nanoparticle substrates was applied to detect and classify PS and PET contamination in four green tea powder varieties across the 400-1650 cm(-1) spectral range. To classify the samples, we compared two chemometric techniques-Partial Least Squares Discriminant Analysis (PLS-DA) and Support Vector Machine (SVM)-with a deep learning approach namely one-dimensional convolutional neural network (1D-CNN). Using optimized preprocessing, PLS-DA achieved perfect classification accuracy (100 %) for all tea varieties except Ryokucha. SVM also showed strong performance but with slightly reduced accuracies of 83.89 % for Matcha, 100 % for Jasmine, and 93.24 % for Sencha. Although the 1D-CNN model achieved higher validation accuracies-93.52 % (Matcha), 99.91 % (Jasmine), and 94.26 % (Sencha)-than SVM, its performance was still slightly lower compared to the PLS-DA models. Additionally, for unknown samples from distinct green tea varieties, the SERS-PLS-DA approach again delivered the highest validation accuracies of 100 % (Matcha), 99.80 % (Jasmine), and 96.46 % (Sencha), further demonstrating the strong potential of this technique. These findings confirm that SERS with gold nanoparticle substrates, especially when integrated with PLS-DA, provides a highly sensitive, non-destructive, and reliable platform for the rapid detection and classification of microplastic contamination in green tea.