Implementation of an open source algorithm for particle recognition and morphological characterisation for microplastic analysis by means of Raman microspectroscopy

Microplastic (MP, i.e. synthetic polymer particles of 1 mu m-5 mm) is not only a suspected environmental contaminant, but also a challenging target for chemical analysis. To analyse particularly very small MP particles, Raman microspectroscopy (RM) enables morphological characterization and chemical identification at the single particle level. To this end, the RM procedure consists of the three steps: particle detection (recognition of particles and their morphological characterization), RM measurements and spectra evaluation. To enable effective and unbiased particle detection on filter samples even when covered by a multitude of particles, we present the implementation of Otsu's algorithm and a watershed-called transformation available as open source code for ImageJ on a common Raman microscope. Otsu's algorithm is an automatic thresholding algorithm that splits pixels in two groups (bright and dark) by minimizing the between-class variance of the two groups. The additional watershed transformation finds the watersheds of particle agglomerations. We demonstrate the effectiveness of our implementation of these algorithms. A coloured microscopic picture is converted into a black and white (b/w) image. Indents (neck-positions) of this image are then used to separate agglomerated particles. This implementation was critically evaluated regarding the criteria accuracy (validity), reliability (precision) and sensitivity. The algorithm presented here (; https://gitlab.lrz.de/raman-sem-iwc/mipran) allows for reliable detection of particles in a wide variety of particle characteristics (size, shape, colour and transparency) and illumination types (dark and bright field). It has the additional benefit that it is equally applicable to any method using image-based particle detection so that it can help in harmonizing MP research throughout the community.