Automatic quantification and classification of microplastics in scanning electron micrographs via deep learning

• DOI: 10.1016/j.scitotenv.2022.153903
• Publication Date: 2022-02-19T15:11:48Z
• Authors: Bin Shi; Medhavi Patel; Dian Yu; Jihui Yan; Zhengyu Li; David Petriw; Thomas Pruyn; Kelsey Smyth; Elodie Passeport; R.J. Dwayne Miller; Jane Y. Howe
• Keywords: Microscopy; Deep Learning; Microplastics; Electrons; Plastics; 01 natural sciences; 0105 earth and related environmental sciences

Microplastics quantification and classification are demanding jobs to monitor microplastic pollution evaluate the potential health risks. In this paper, microplastics from daily supplies in diverse chemical compositions shapes imaged by scanning electron microscopy. It offers a greater depth finer details of at wider range magnification than visible light microscopy or digital camera, permits further composition analysis. However, it is labour-intensive manually extract micrographs, especially for small particles thin fibres. A deep learning approach facilitates with annotated dataset including 237 micrographs (fragments beads) 50 μm-1 mm fibres diameters around 10 μm. For quantification, two models (U-Net MultiResUNet) were implemented semantic segmentation. Both significantly outmatched conventional computer vision techniques achieved high average Jaccard index over 0.75. Especially, U-Net was combined object-aware pixel embedding perform instance segmentation on densely packed tangled quantification. shape classification, fine-tuned VGG16 neural network classifies based their accuracy 98.33%. With trained models, takes only seconds segment classify new micrograph accuracy, which remarkably cheaper faster manual labour. The growing datasets may benefit identification environmental samples future work.