Paint may be doing far more than adding color to buildings, cars, ships and roads: as coatings weather and peel, they can become a major source of microplastic pollution. A new study has assembled what researchers describe as the first dedicated toolkit for recognizing these fragments, combining a spectral library, a visual identification guide and a machine-learning model. The resource, called the Paint Library of Plastic Particles, or PLoPP, is designed to help scientists distinguish paint-derived particles from other forms of plastic debris that can look almost identical under a microscope. The work addresses a problem that has quietly complicated microplastic surveys for years. Many environmental particles are too small, degraded or chemically complex to identify reliably by appearance alone, yet paint fragments may carry distinctive chemical signatures that can reveal where they came from and how they move through ecosystems.

Paint is a composite material rather than a single type of plastic. Modern coatings can contain polymers, pigments, binders, additives and mineral fillers, and their formulations vary widely depending on whether the product is intended for a house, automobile, ship, road surface, industrial structure or wooden object. Sunlight, heat, abrasion, salt water and repeated freezing and thawing can gradually break down a coating into particles ranging from visible flakes to microscopic fragments. Once released, those particles can enter stormwater, rivers, coastal sediments and the ocean. Some may be carried through the atmosphere as dust, while others accumulate near heavily painted infrastructure, harbors and roadways. Because the particles often retain colors and textures associated with their original coatings, they may be recognizable to a trained observer—but visual clues become less dependable as fragments weather, lose pigment or become mixed with other debris.

To build PLoPP, the researchers analyzed 90 paints spanning seven sectors: architectural, automotive, consumer, general industrial, marine, road-marking and wood coatings. The collection covered 15 colors and five appearances, including glitter, gloss, matte, pearl and semi-gloss finishes. It also represented at least 25 polymers, although polyurethane, polyurethane acrylics and polyvinyl chloride dominated the library. From these materials, the team generated 263 spectra. A spectrum is effectively a chemical fingerprint: it records how a material absorbs infrared light at different wavelengths. Different molecular bonds vibrate at characteristic frequencies, allowing scientists to infer the chemical composition of a tiny particle even when its origin cannot be established by sight. By assembling many reference spectra in one paint-specific database, the researchers aimed to give environmental scientists a much stronger comparison set than a general plastic library could provide.

The central analytical technique was attenuated total reflectance Fourier-transform infrared microspectroscopy, known as µATR-FTIR. In FTIR analysis, infrared radiation is directed at a sample and the resulting pattern of absorbed wavelengths is measured. The attenuated-total-reflectance approach uses contact between the sample and a crystal to probe the material’s surface, while microscopy allows the instrument to target individual particles rather than a bulk mixture. That distinction matters because environmental samples commonly contain many particle types at once. A paint flake may be only one item among fibers, packaging fragments, tire-related particles, biological material and mineral grains. A general FTIR database can identify the polymer class, but it may not distinguish a painted plastic fragment from an unpainted fragment made from a similar polymer. PLoPP adds paint-specific reference patterns that can improve that decision.

The researchers also created a visual key to make identification possible before, or alongside, instrumental analysis. The guide organizes particles according to observable traits such as color, surface appearance and morphology. A fragment with a bright metallic sheen, layered structure or a characteristic matte surface may provide an immediate clue that it originated from a coating. Yet the study’s design recognizes that appearance is not proof of composition. Weathering can make glossy particles dull, while pigments and additives can obscure the underlying polymer signal. The visual key therefore works as a structured screening method rather than a replacement for spectroscopy. Its value is particularly important for laboratories that do not have immediate access to advanced instruments, and it may help researchers select which particles deserve more detailed chemical analysis.

To test whether the reference collection could separate paint from other microplastics, the team developed a spectral-analysis pipeline using a support vector machine, a type of machine-learning algorithm commonly used to classify complex data. The model learns boundaries between categories by examining patterns in the spectra rather than relying on a single chemical peak. Preprocessing steps included standard normal variate transformations, which can reduce variation caused by scattering and differences in signal intensity, and principal component analysis, which compresses many correlated spectral measurements into a smaller number of meaningful dimensions. When tested on pristine paint and non-paint microplastic samples, the model achieved an overall accuracy of 92 percent. That result indicates that the chemical fingerprints contained enough information to distinguish the two groups under controlled conditions.

Environmental samples presented a more difficult challenge, as the researchers expected. Using particles collected in Plymouth, United Kingdom, and spectra from Charleston, South Carolina, the team examined how the tools performed on materials that had been exposed to real-world conditions. The machine-learning model’s accuracy fell to 55 percent when it attempted to differentiate environmental paint particles from non-paint microplastics. Weathering likely contributed to the decline: ultraviolet radiation, oxidation, abrasion and chemical exposure can alter the surface chemistry of a particle, while dirt and biological films can add signals not present in pristine samples. The lower result is an important warning against treating laboratory accuracy as a direct measure of field performance. A model trained on clean reference materials may need much broader training data before it can reliably classify the chemically messy particles found in nature.

The other approaches performed better in the environmental tests. The visual key achieved an average accuracy of 92 percent for particles, while correlation-based searches using PLoPP in OMNIC software correctly classified 86 percent of environmental particle spectra as paint or non-paint. Correlation-based searching compares the shape of an unknown spectrum with reference spectra and assigns a match according to their similarity, often expressed through a hit quality index. Together, these findings suggest that no single method is likely to solve the identification problem in every setting. Visual assessment can be fast and surprisingly effective, but it depends on training and may be vulnerable to observer judgment. Spectral searches provide chemical evidence, though results can be affected by weathering and mixed materials. Machine learning can process large numbers of spectra, but its reliability depends heavily on how representative its training library is.

The researchers say PLoPP is intended as a foundation for improving estimates of paint-related microplastic pollution, not as a final classification system. One unanswered question is whether paint fragments can be assigned reliably to the sector in which they were used. If spectra or combinations of pigments, polymers and surface features prove distinctive enough, future versions of the library might help connect particles to road markings, marine coatings, buildings or vehicles. That could allow scientists to identify pollution hotspots and determine which activities contribute most to environmental contamination. For now, the study demonstrates the practical value of creating a paint-specific reference collection. By making the hidden fingerprints of coatings easier to recognize, the researchers offer a way to bring a previously overlooked source of microplastics into sharper focus—and potentially transform how scientists track the colorful fragments accumulating beyond the surfaces they once protected.

Subject of Research: Identification of paint-derived microplastic particles using infrared spectroscopy, visual classification, and machine learning

Subject of Research: Technology and Engineering

Article Title: A Paint Library of Plastic Particles (PLoPP): a spectral library, visual key, and machine learning model for paint microplastic identification

Article References: Diana, Z. T., Ford, J., Rubinovitz, R., Turner, A., Milne, M. H., & Rochman, C. M. (2026). A Paint Library of Plastic Particles (PLoPP): a spectral library, visual key, and machine learning model for paint microplastic identification. Microplastics and Nanoplastics. https://doi.org/10.1186/s43591-026-00222-4

Image Credits: AI Generated

DOI: 10.1186/s43591-026-00222-4

Keywords: paint microplastics, microplastic identification, FTIR spectroscopy, spectral library, machine learning, environmental particles, plastic pollution, visual classification

Cite Scienmag News
APA MLA Chicago

Everett F. (August 29, 2026). PLoPP: Spectral Library and Machine Learning Identify Paint Microplastics. Scienmag. https://scienmag.com/plopp-spectral-library-and-machine-learning-identify-paint-microplastics/

Copy citation Download RIS

Tags: chemical signatures of paint debrischemical signatures of paint particlescoatings degradation and microplastic releaseecosystem contamination by paint particlesenvironmental impact of paint microplasticsenvironmental microplastic source trackingenvironmental microplastic surveysmachine learning in environmental analysismachine learning in environmental sciencemicroplastic debris analysismicroplastic pollution detectionmicroplastic pollution identificationmicroplastic survey challengespaint fragment detection in ecosystemspaint fragment identification toolkitpaint-derived plastic particle recognitionpaint-derived plastic particlespolymer composition in paint microplasticspolymer composition of paint particlesspectral analysis of microplasticsspectral library for paint microplasticsvisual identification of microplastic fragmentsvisual identification of microplastics