HMPD: A Novel Dataset for Microplastics Classification with Digital Holography

Microfiber shedding caused by the washing of synthetic textiles is widely recognized as a major source of environmental microplastic pollution. Such a phenomenon pushed the scientific community to develop ever better methodologies for the detection and the identification of synthetic microfibers dispersed in water involving, mainly because of the microscopic nature of the elements under analysis, a wide range of competencies ranging from instrumental acquisition to data analysis. This work deploys a large dataset of holographic images of microplastic and non-microplastic samples carefully labeled by experts. Data acquisition starts from video sequences collected by means of a holographic microscope recording the flow of water samples; then, holography post-processing and object detection are performed to retrieve the patches of interest. Finally, classification tests exploiting deep learning strategies have been performed to validate the proposed dataset. The dataset and the benchmarking code are available at: https://github.com/beppe2hd/HMPD.