The framework operates at three nested levels. At the image level, the team used latent variable selection to handle variation arising when acquisition conditions change between measurements. At the tile level, a background-based correction reduced illumination-related artefacts that can distort chemical maps. At the spectral level, preprocessing combined baseline correction, smoothing, derivative calculation, normalization, and wavelength selection, and only spectra belonging to particles were retained for further analysis, trimming the computational load considerably.

For polymer identification, the researchers introduced a scalable strategy built on clustering. Rather than matching every individual pixel spectrum against a reference library, particle spectra were first grouped into clusters, and library matching was performed on cluster centroids. Among twelve matching strategies evaluated, a sign-invariant derivative-based cosine similarity method achieved perfect classification accuracy for four common polymers: polystyrene, polyethylene terephthalate, polyethylene, and polypropylene.

The clustering-based workflow also produced more spatially coherent particle maps than direct software-based matching, meaning identified regions appeared as cleaner, more contiguous particle shapes rather than scattered pixels. Processing time was substantially reduced compared with the direct approach, a gain the authors attribute to matching centroids instead of potentially thousands of individual spectra per image.

The framework was evaluated for supervised classification-based microplastic identification, and the reported accuracy figures apply to the polymers and data tested in the study. According to the authors, the combination of multi-level correction with cluster-centroid spectral matching improves the robustness, efficiency, and interpretability of spectral-imaging-based microplastic identification, offering a route to handling the large data volumes and artefacts that have limited routine application of the technique.

Subject of Research: A multi-level preprocessing and modelling framework for FT-IR spectral imaging of microplastics

Article Title: A multi-level preprocessing and modelling framework for spectral imaging of microplastics

Article References: Duma, Z.-S., Tsering, T., Heikkinen, S., Soininen, T., Sihvonen, T., Koistinen, A., & Reinikainen, S.-P. (2026). A multi-level preprocessing and modelling framework for spectral imaging of microplastics. Microplastics and Nanoplastics. https://doi.org/10.1186/s43591-026-00242-0

Image Credits: AI Generated

DOI: 10.1186/s43591-026-00242-0

Keywords: microplastics, FT-IR spectroscopy, spectral imaging, preprocessing, clustering, polymer identification, cosine similarity, polystyrene, polyethylene, polypropylene, PET, data analysis