Related Experiment Video
Updated: Sep 2, 2026
Sampling, Sorting, and Characterizing Microplastics in Aquatic Environments with High Suspended Sediment Loads and Large Floating Debris
Published on: July 28, 2018
Portable Raman spectroscopy coupled with machine learning for rapid identification and source apportionment of
Jing Lin1,2, Panting Ren1,3, Zhenglong Chen1
1College of Chemistry and Materials Science, Fujian Provincial Key Laboratory of Advanced Oriented Chemical Engineer, Fujian Key Laboratory of Polymer Materials, Engineering Research Center of Industrial Biocatalysis, Fujian Province Higher Education Institutes, Fujian Normal University, Fuzhou, Fujian 350007, China.
Abstract:
Aquaculture expansion has exacerbated microplastics (MPs) contamination in aquaculture water bodies. MPs readily adsorb heavy metals and organic pollutants to form composite contamination and accumulate through food chains, imposing ecological and human health risks. Conventional detection techniques including microscopic observation, FTIR and Py-GC-MS are limited by low identification accuracy, matrix interference, destructiveness and cumbersome procedures, which cannot support rapid large-scale monitoring. Raman spectroscopy enables non-destructive detection with unique molecular fingerprinting, yet spectral overlap, background noise and subtle crystallinity differences among similar plastics hinder accurate manual classification. In this study, Raman spectroscopy was integrated with machine learning to establish a rapid plastic particle classification approach. Raman spectra of thirteen typical standard MPs were acquired, preprocessed and dimensionally reduced by PCA-LDA. Seven machine learning models were constructed and optimized by cross-validation, and further validated using spiked aquaculture wastewater samples. All models achieved classification accuracies above 98%. Among them, K-nearest neighbor (KNN), naive Bayes (NB), support vector machine (SVM), and logistic regression (LR) exhibited superior classification performance due to their effective feature discrimination capability and adaptability to high-dimensional Raman spectral data, enabling accurate classification of plastic particles according to polymer types under complex aquatic matrices. The proposed method integrates the molecular fingerprinting capability of Raman spectroscopy with the feature-learning advantages of machine learning, effectively overcoming the limitations of conventional spectral interpretation. This portable and non-destructive strategy provides a reliable approach for rapid plastic particle classification, pollution source tracing, and ecological risk assessment in aquatic environments.
