Imaging-based lensless polarisation-resolving fluid stream analyser for automated, label-free and cost-effective microplastic classification

The presence of microplastics in the environment is of concern with the actual distribu-tion of this pollution remaining relatively unknown. The ocean is of particular interest as the monitoring of microplastics in this area presents a challenge in that in situ fluid stream solutions are not readily available and traditional sampling methods are labour-intensive and costly. Additionally, the lack of consensus on sampling techniques makes comparing results dicult. Our proposed device demonstrates an imaging-based lens-less polarisation-sensitive fluid stream analyser (FSA) for automated, label-free, and cost-e↵ective microplastic classification. The FSA performs analysis at high flow rates with a custom-designed illumination circuit that reduces motion blur and provides quan-titative sample information using a polarisation-sensitive image sensor. Digital in-line holography (DIH) and birefringence numerical computation are utilised in the processing workflow. The device can be used for either quantitative polarisation-sensitive imaging and analysis or for further machine-learning-based activities, including the classification of samples. Both abilities are demonstrated in this study. Our analyser computes the two-dimensional birefringent characteristics of samples and we investigate the detection of synthetic polymer birefringent textures due to the optical anisotropy of these materi-als. We perform a comparative machine learning study with both learned and filter bank feature generation being assessed to aid the microplastic classification process. The FSA and classifier components are used to develop an end-to-end workflow that samples a fluid stream and determines the composition of marine and microplastic particles. We use two phytoplankton cultures to create a simplified marine environment for testing purposes. To demonstrate the performance of our classification methods we tested our device and workflow in a two-class configuration for marine microorganisms and plastics, as well as a five-class configuration for marine microorganisms and four individual plastic types (polyethylene (PE), polyethylene terephthalate (PET), polypropylene (PP), and polystyrene (PS)). Our analysis shows that high accuracy is achieved from the classifier implementation, with the simulated marine environment experiments further supporting the ability of the proposed implementation.