Single particle identification and automated classification of small (<10µm) microplastics using cathodoluminescence

Despite growing awareness of the negative impacts of plastic pollution, its production and resulting waste continue to increase. Once in the environment, plastic waste breaks down into microplastic (MP) particles, which have been detected in various environmental and biological matrices, including air, water, sediment, blood, and brain tissue. However, the full extent of the problem remains unclear due to limitations in current detection techniques. This paper builds on our previous work with Scanning Electron Microscopy coupled with Cathodoluminescence (SEM-CL) and demonstrated that small particles of commonly used plastics exhibit unique CL spectra. After optimizing signal parameters, we successfully collected over 200 CL spectra of test materials (1-10 µm) of polyethylene (PE), polypropylene (PP), polyamide (PA), polystyrene (PS), and polyethylene terephthalate (PET). Principal component analysis (PCA) confirmed the uniqueness of CL spectra for differentiating plastic types. A random forest classifier (RFC) was trained on the first 10 principal components (PCs) and achieved 94% accuracy. To minimize pre-processing, a second RFC was trained on normalised CL spectra, also achieving 94% accuracy. By applying a certainty threshold, untrained contaminants such as kaolin, talc, and titanium dioxide were separated from the plastics. This work laid the foundation for a robust detection tool for single particle identification and automated classification of microplastics smaller than 10 µm using SEM-CL. • Successful collection of CL spectra of environmentally relevant microplastic particles with sizes below 10 µm • RFC classifier build with an accuracy of 94% for microplastic particles • Application of a certainty threshold to distinguish between target particles and contaminants