A Microwave-Based Sensing Platform for Microplastic Detection and Quantification: A Machine Learning-Assisted Approach

This work introduces a low-cost microwave (MW) sensor that achieves the highest sensitivity among MW-based approaches for detecting microplastics (MPs) in water and provides a mechanism for determining polymer types in unknown samples. These achievements address the growing need for sensitive, affordable tools for routine and reliable monitoring of MPs across environmental water matrices. We employ a MW spiral resonator operating at 3.4 GHz to detect and quantify polytetrafluoroethylene (PTFE), polyvinyl chloride (PVC), and polyethylene terephthalate (PET) in water. The design leverages a highly sensitive spoof surface plasmon polariton (SSPP) mode and a controlled field-sample interaction, enabling tunable coupling and robustness to variations in sample composition. Under maximum interaction (1.6 mL sample volume), the resonator detects concentrations as low as 125 ppm for all polymers. To overcome overlapping responses from similar dielectric properties, we integrate machine learning (ML) to discriminate polymer type. Four algorithms, Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Logistic Regression (LR), were evaluated using resonance-derived features: frequency shift, minimum amplitude, 3 dB bandwidth, and skewness. Logistic Regression achieved the best performance, correctly classifying 100 % of samples. After identification, polymer-specific calibration curves are used to estimate the concentration. These findings establish the platform as a sensitive, selective, and low-cost solution for MP monitoring in water. By coupling SSPP-enabled sensing with ML-based classification, this work advances the state-of-the-art in MW environmental monitoring and offers a practical route for addressing MP pollution while enabling both quantification and identification in a single device.