We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Self-organizing maps for unsupervised microplastic detection using Raman hyperspectral images
Summary
Scientists developed a smarter computer method that uses light-scattering patterns (Raman spectroscopy) to automatically spot microplastic particles in water samples, without needing pre-programmed labels or training data. This matters because it could make testing our water for microplastics faster, cheaper, and more accurate—an important step toward better understanding and monitoring the tiny plastic particles we may be ingesting daily. The method even worked in real river water, suggesting it could be practical for real-world environmental monitoring, though more research is needed before it's used to directly assess health risks.
As microplastic contamination in the environment continues to increase, the development of accurate, label-free detection methods remains a bottleneck for routine analytical workflows. In this work, we present an unsupervised chemometric workflow implemented in the Google Colaboratory Python environment that uses self-organizing maps (SOM) to produce topology-preserving representations, perform clustering, and visualize Raman hyperspectral images of microplastics. Raman hyperspectral images were obtained after filtering polysulfone microparticles, and the data were processed using a two-dimensional SOM trained on a hexagonal lattice, which supports spatially coherent pixel-level segmentation and topographic error values down to 0.087. SOM prototype vectors captured local polysulfone-like spectral motifs, with cosine similarity values up to 0.94. Spatial reconstruction of the images recovered particle-containing regions for single and multiple particles (from 37/900 to 160/900 pixels and from 7/144 to 23/144 SOM neurons). The proposed method demonstrated advantages over univariate and principal component analysis approaches and was successfully applied to a river water matrix. Excellent chemical interpretability was also evidenced by component-plane analysis, which highlighted the C–O–C stretching band of polysulfone at 1142 cm -1 and background-related contributions at 883 and 1232 cm -1 . Therefore, the proposed SOM workflow may be readily implemented as a spectroscopically interpretable approach to microplastic detection in complex sample matrices.