Recently, a research team led by Prof. GAO Xiaoming and LIU Kun from the Hefei Institutes of Physical Science of the Chinese Academy of Sciences, developed a cascaded neural network for analyzing mixed microplastic Raman spectra.
Their findings were recently published in Analytical Chemistry.
Microplastics, which are smaller than 5 mm, are widespread and pose growing environmental and health concerns. Raman spectroscopy combined with neural networks offers a promising solution for their identification, but accuracy is often limited by overlapping signals in complex samples and low quality data collected in real world environments.
The new approach used a cascaded neural network framework that was capable of robust spectral reconstruction, efficient classification, and effective unmixing of overlapping microplastic signals, even under these challenging measurement conditions.
Testing on about 20–25 mixed microplastic samples collected under varying conditions, the framework demonstrated remarkable improvements. When measured under low laser power (~50 mW) and short integration time (~3 seconds), classification accuracy jumped from a moderate 52% to a high 91%, showing its effectiveness in situations where traditional methods struggled.
Key to this success is the integration of a channel and spatial attention module, which enhances feature extraction without adding heavy computational cost, and a dynamic hybrid physical loss function, which accelerates training and improves model convergence.
This work provides a promising solution for accurate microplastic identification in complex and low-quality measurement environments, supporting more reliable environmental monitoring and research.
The Raman spectra of detected microplastics are input into a neural network trained to output clean Raman spectra. Spatial and channel attention modules enhance the model' s performance. Training is optimized using a custom loss function. Grad-CAM visualizations highlight the regions of the spectra that the neural network identifies as significant. (Image by HUANG Weixiang)