Dual-Channel Fluorescence Sensor Sorts Microplastics at 77.8% Accuracy

A proof-of-concept optical sensor has classified six common microplastic polymers in flowing water with 77.8% accuracy, using a 405 nm laser at just 4.5 mW and a solvatofluorochromic dye. The system, named SMILE (Sensing of Microplastics in Liquid Environments), fits on a 45 × 30 cm portable aluminum breadboard and analyzes samples at flow rates up to 600 mL/h. The findings appear in Scientific Reports, based on an accepted but not yet final-edited paper.

That performance number matters because it comes from continuous flow measurement, not from bench spectroscopy. Microplastic particles have shown up in up to 83% of global tap water samples, yet the reference methods — Fourier-transform infrared (FTIR) spectroscopy and Raman microspectroscopy — need expensive instruments and slow, multi-stage sample prep that makes continuous monitoring impractical. Existing flow cytometers and fluorescent-dye counters can tally particles, but dyes like Nile Red can't reliably tell polymers apart. In short, the field can count the problem but can't name it in real time.

How does the sensor work?

The physics hinge on solvatofluorochromism: the dye 4-dimethylamino-4'-nitrostilbene (DANS) shifts its fluorescence emission wavelength depending on the polarity of its chemical surroundings. Different polymers present different polarities, so each shifts the dye's emission differently.

The hardware chain is straightforward. An analog pump pushes the sample through a glass capillary with a 200 μm inner-diameter constriction at 430 mL/h. A 49 μm inline metal filter upstream blocks larger debris. A dichroic mirror steers the 405 nm excitation beam onto the constriction, forming an elliptical focal spot of 200 × 50 μm. The same lens collects the emitted fluorescence, and a secondary beam splitter divides it into two spectral channels: CH1 for wavelengths above 550 nm (polar materials, yellow-to-red) and CH2 for below 550 nm (non-polar polymers, blue-to-green). Each channel feeds its own photomultiplier tube.

A 16-bit data acquisition board digitizes the pulses, and custom MATLAB code handles baseline correction, coincidence detection across both channels, and noise rejection via pulse-duration thresholds of 10–120 μs. Classification comes from the ratio of integrated pulse areas between the two channels — a ratiometric approach that suppresses common optical noise sources.

The takeaway: the polymer ID comes from a ratio between two photomultiplier channels, not from a full spectrum.

Which polymers did it separate cleanly?

The polarity gradient shows up clearly in the calibration data. Non-polar polypropylene and polyethylene emitted peaks near 480 nm, landing mostly in CH2 — polypropylene at a reference angle of 90.0°, polyethylene with a fitted slope of 18.2 (86.9°). Polystyrene sat in between, peaking near 530 nm with a slope of 5.77 (80.2°). PMMA, PVC, and PET emitted between 560 and 580 nm, registering stronger in CH1, with fitted slopes of 2.79, 2.40, and 1.84 respectively. Independent micro-Raman spectroscopy confirmed the polymer identities.

Fivefold cross-validation on 816 detected particle events produced the 77.8% headline accuracy. Recall tells a more nuanced story:

• Polystyrene: 98.6%
• Polyethylene: 95.4%
• Polypropylene: 93.2%
• PET: 59.6%
• PMMA: 41.1%
• PVC: 24.5%

The polar polymers suffered from overlapping fluorescence responses. The authors also flag limited particle-count recovery, from 40% for PVC to 65% for PET, relative to nominal suspension concentrations — so the counting side needs work before environmental deployment.

Non-polar polymers classify near-perfectly; the polar trio still confuses the algorithm.

Where would it actually be used?

The target environments are municipal drinking water facilities, filtration plants, and industrial wastewater systems. Because SMILE analyzes particles directly in flowing water at hundreds of milliliters per hour, it eliminates substrate transfers that risk particle loss or contamination. The authors see inline deployment across water intake facilities and distribution networks, providing early warning of concentration spikes and acute contamination events.

Next steps include testing weathered, degraded, and biofouled polymers — real environmental particles, not clean lab spheres — adding optical channels to sharpen classification, and applying machine learning to improve discrimination and extract particle size.

The open question for water utilities: is 77.8% accuracy with sub-65% count recovery enough for screening duty, or does the technique stay in the lab until the polar-polymer overlap is resolved?

Journal reference: Nicolai, E., et al. (2026). A dual-channel optical sensor for real-time microplastic classification via ratiometric fluorescence analysis. Scientific Reports. DOI: 10.1038/s41598-026-78675-z.

By Sophie Lindqvist