Identification and visualization of environmental microplastics by Raman imaging based on hyperspectral unmixing coupled machine learning

Abstract

Microplastics (MPs) are ubiquitous contaminants that have become an emerging pollutant of concern, potentially threatening human health and ecosystem environments. Although current detection methods can accurately identify various types of MPs, it remains necessary to develop non-destructive and rapid methods to meet growing demands for detection. Herein, we combine a hyperspectral unmixing method and machine learning to analyse Raman imaging data of environmental MPs. Five MPs types including poly(butylene adipate-co-terephthalate) (PBAT), poly(butylene succinate) (PBS), p-polyethylene (PE), polystyrene (PS) and polypropylene (PP) were visualized and identified. Individual or mixed pure or aged MPs along with environmental samples were analysed by Raman imaging. Alternating volume maximization (AVmax) combined with unconstrained least squares (UCLS) method estimated end members and abundance maps of each of the MPs in the samples. Pearson correlation coefficients (r) were used as the evaluation index; the results showed that there is a high similarity between the raw spectra and the average spectra calculated by AVmax. This indicates that Raman imaging based on machine learning and hyperspectral unmixing is a novel imaging analysis method that can directly identify and visualize MPs in the environment.