Classification of (micro)plastics using cathodoluminescence and machine learning

Abstract

Microplastics are a growing environmental and toxicological concern, having been found in the remotest locations of the earth and within multiple organs of the human body. However, the scale of the problem is not yet fully known as the smallest micro- and nanoplastics (MNPs) cannot accurately be measured due to limitations in measurement and detection techniques. In this paper, we combine the nanoscale resolution of Scanning Electron Microscopy (SEM) with the spectroscopic power of cathodoluminescence (CL) to show that six of the most common plastics – HDPE, LDPE, PP, PA, PS and PET – have unique spectra that could help in identifying the smallest MNPs. This was done by building a spectral database using 111 plastic samples from reference and consumer plastics with different sizes (0.001–1 mm), colours (e.g. black, blue, green, red, white/transparent and yellow) and shapes (e.g. irregular, fibre, spheres). We then trained multiple classification models using an Artificial Neural Network approach. With the use of these classification models we were able to classify the six plastics, including difficult samples such as black coloured plastics, based on their CL spectra with 97% accuracy, showing that our approach is robust towards sample differences. As most misclassifications occurred between LDPE and HDPE, a separate model for LDPE and HDPE allowed for >99% accuracy for the classification of HDPE and LDPE by using a two-step approach. This novel “proof-of-concept” to MNP analysis demonstrates the utility of SEM paired with CL to characterise microplastics with detailed spatial and chemical resolution warranting its further development and adoption.