Automatic microfibers detection in SEM images

Microplastics, plastic debris with particles size ranging from 1 nm to 5 mm, have a negative health impacts on marine life, resulting in environmental and human health threats. The fiber microplastics are considered one of the major sources of plastic pollution in oceans, probably due to the fact that, because of their shape, they can partially pass-through wastewater treatment plants and reach directly the sea. Microfibers (MFs) pollution come from a variety of sources such as synthetic fishing nets and ropes, but mainly from textiles and their wastewater. MFs produced by washing processes of synthetic textiles is mainly due to the mechanical and chemical stresses due to washing cycles, detergents, temperature and treatment washing time. The existing analytical methods for microplastic quantification range from visual identification to molecular spectroscopy. In the microscopy field, the use of a Scanning Electron Microscope (SEM) offers extremely clear and high-resolution images of plastic particles, but related image analysis has proven more challenging than in traditional optical microscopy. Unaided visual identification and count of microfibers require skilled and experienced operators and results to be a lowthroughput and subjective method. The development and application of automated techniques for extracting relevant picture features and support the final human counting has therefore become necessary. A large AI model, applied for fibers detection and counting in SEM images, is presented here. It is based on Zero-shot learning, an approach that leverages the deductive capabilities of models, making it possible to be applied, for several tasks of fibers counting in SEM images, without human intervention. The proposed approach is tested and validated on a set of micrographs of wastewater from four washing cycles of cotton fabrics.