Deep learning vs classical computer vision techniques for microplastics classification

spectroscopy or Raman scattering spectroscopy are commonly used, thanks to their complementarities.Some vibrational modes are more active in a technique than in the other depending on the molecule symmetry and on chemical groups.To generate reliable results, the comparison with reference spectra database is absolutely necessary to unambiguously identify polymer type.Given the large number of spectra to analyze and since the spectra libraries usually consist of spectra of pure substances, thus spectra obtained from environmental samples are expected to have low congruity compared to reference spectra.The current solution does not allow easy assignment and fast identification of particles.In order to extend this identification, an alter-native solution is based on multivariate analysis.The spectra were analyzed using Independent Component Analysis (ICA).It consists to separate multivariate signal into subcomponents, sup-posing the mutual statistical independence of the non-Gaussian source signals.This method of identifying microplastic particles using multivariate methods is very powerful, as it takes into account the whole spectrum.This method helps to identify particles type by identifying copoly-mers and plasticizers, and to distinguish plastic particles and fibers from non-plastics.Those approaches will be presented on both Raman scattering and FTIR techniques, and perspectives on fast microplastic identification will be discussed.