Environmental degradation of consumer plastics into microplastics and nanoplastics and their classification using machine learning

Consumer plastics such as plastic shopping bags, water bottles and take-food trays can be sources of secondary microplastics and nanoplastics (MNPs) when they undergo complex environmental degradation. Under the influence of UV radiation, abrasion, atmospheric oxidation and freeze-thaw cycles, bulk plastics degrade into smaller particles with significantly increased transportation in the aquatic environment. In this work, we studied the degradation of three classes of consumer plastics, namely, plastic shopping bags (low density polyethylene), water bottles (polyethylene terephthalate) and take-out food trays (polystyrene) under simulated environmental condition. The first MNPs emerged after at least 4 weeks of simulated environmental degradation. We characterized the transformation of the surface chemistry of plastics using Fourier transform infrared spectroscopy (FTIR) and X-ray photoelectron spectroscopy (XPS). Our results show that environmental degradation has a significant effect on alteration of the surface chemistry of polymers by inducing a plethora of oxygen containing groups. The morphology of the MNPs and post-degradation bulk plastics were also studied by scanning electron microscopy (SEM). A dataset of 101 annotated micrographs was generated for comprehensive data analysis. The dataset was structured to accommodate various applications, including MNP detection, categorization by polymer type, and classification based on size. The images were manually annotated, including masks, bounding boxes, area, diameter, and category information for each particle. The dataset exhibited diverse particle density and size distributions, with varying concentrations of particles across different images. The dataset's statistics encompassed the distribution of the number of particles in images and the distribution of particles according to particle diameter. Cutting-edge algorithms, such as U-net for semantic segmentation and Mask-RCNN for instance detection, were applied to the dataset for validation. These experiments highlight the dataset's value in addressing open problems in MNP and particle detection, particularly in scenarios with overlapping particles, crowded scenes, and clumped particles. Also see: https://micro2024.sciencesconf.org/559714/document