AI-Powered Fusion Model Improves Detection of Microplastics in the Atmosphere

In a recent study, a team of researchers from Nanjing University of Information Science and Technology examined a new approach that could improve environmental pollution control. This study, which was published in the journal Spectrochimica Acta Part B: Atomic Spectroscopy, presented a new method that can accurately monitor atmospheric microplastics (1). This novel method involved the use of an image-led and spectroscopy-assisted fusion system.

What are microplastics, and what threat do they pose for the environment?

Microplastics (MPs) are tiny fragments of plastic that are no larger than 5 mm (2). Currently, several studies have examined ways to better monitor MPs in the environment (2–4). Apart from their negative environmental impact, MPs also pose health risks because they can adsorb hazardous chemicals, including heavy metals such as lead (Pb) and chromium (Cr) (1).

What did the researchers test in their study?

In their study, the researchers used an image-led and spectroscopy-assisted fusion system to rapidly and accurately detect and quantify toxic contaminants on polyamide (PA) MPs suspended in the air. The main objective was to test whether their method could perform this task accurately and efficiently.

One of the key aspects to their study was how the researchers incorporated machine learning (ML) into their method. Using maximum variance convolutional neural network (MV-CNN), the researchers improved the recognition of MP particles with adsorbed heavy metals (1). MV-CNN operated differently from traditional CNNs by incorporating spatial variance maximization and principal component analysis (PCA) to prioritize the important image features while reducing redundancy in high-dimensional data sets (1). This operational adjustment helped the researchers improve classification accuracy. Normally, traditional CNNs achieve approximately a 75% accuracy rate, but MV-CNN was able to reach a 91.67% classification accuracy rate (1).

The MV-CNN was trained on 400 labeled image samples, which by ML standards is a small data set. The results demonstrated its potential in real-world applications where labeled data is limited.

What differentiates this study from previous works?

The researchers tried something new in their study. They created a synergistic “image-led, spectrum-assisted” multimodal strategy by using laser-induced breakdown spectroscopy (LIBS) and digital microscopy imagery (1). LIBS was applied to extract fingerprint spectral features of lead and chromium on the microplastic surfaces. By merging LIBS spectra with image data, the researchers achieved an improvement in quantitative accuracy. The classification accuracy for heavy metal concentration levels ranging from 200 to 1000 ppm jumped from less than 65% to over 84% (1).

In situ detection of MPs has encountered technical challenges as the technology and analytical instrumentation have improved. The fusion system created by the research team combined digital morphology analysis with elemental fingerprinting to overcome these challenges (1). The simulation experiments also showed that the LIBS technique could effectively detect cyanogen (CN) molecular bands and heavy metal spectral peaks associated with polyamide (PA) microplastics in ambient air samples.

One particularly novel finding was the enhanced spectral signal observed in samples where PA microplastics simultaneously adsorbed both Pb and Cr (1). The bimetallic system showed significantly stronger LIBS signals compared to monometallic counterparts, offering clues to the synergistic adsorption mechanisms at play (1). Changes in color features accompanying the dual adsorption provided key morphological markers that MV-CNN used for more accurate classification (1).

A linear regression model developed from the LIBS data showed a strong correlation (R² > 0.86) between heavy metal concentration and spectral line intensity, further validating the system's potential for quantitative field use (1).

What are the next steps in this work?

The researchers demonstrated that the fusion strategy provides a cost-effective and accurate detection system for environmental monitoring. As a result, it is expected that more portable, real-time sensing devices will be created with the fusion strategy in mind. This technology could be important for air quality regulators, environmental researchers, and public health agencies seeking to mitigate the risks posed by airborne MPs (1).

References

1. Wang, Z.; Aizezi, N.; Ye, Y.; et al. A Multimodal Learning System based on Maximum Variance Convolutional Neural Networks and Laser-Induced Breakdown Spectroscopy for In situ Online Analysis of Heavy Metal Contamination from Atmospheric Polyamide Microplastics. Spectrochimica Acta Part B: At. Spectrosc. 2025, 232, 107270. DOI: 10.1016/j.sab.2025.107270
2. Wetzel, W. Evaluating Microplastic Detection with Fluorescence Microscopy and Raman Spectroscopy. Spectroscopy. Available at: https://www.spectroscopyonline.com/view/evaluating-microplastic-detection-with-fluorescence-microscopy-and-raman-spectroscopy (accessed 2025-07-16).
3. Wetzel, W. New Spectral Method Overcomes Key Barrier in Plastic Pollution Detection. Spectroscopy. Available at: https://www.spectroscopyonline.com/view/new-spectral-method-overcomes-key-barrier-in-plastic-pollution-detection (accessed 2025-07-16).
4. Wetzel, W. Microplastics Found in Deepest Reaches of Central Indian Ocean. Spectroscopy. Available at: https://www.spectroscopyonline.com/view/microplastics-found-in-deepest-reaches-of-central-indian-ocean (accessed 2025-07-16).