Key points are not available for this paper at this time.
The mechanical recycling of plastics is one of the most efficient approaches for reducing carbon dioxide emissions. Purification of the plastics from shredded waste materials requires versatile techniques, such as optical identification by Raman spectroscopy. The identification procedure demands the spectroscopy expertise to assign molecular structures from spectral peaks. In this study, we demonstrate applications to classify plastics using machine learning techniques under practical recycling industry conditions. Combining the techniques of principal component analysis (PCA) and support vector machine provides an accurate and robust classification of the valuable plastics of polypropylene, polystyrene, and acrylonitrile-butadiene-styrene copolymer. The identification accuracy remained above 95%, even with noise 3 times larger than the original intensity. For noise 10 times larger, the accuracy was more than 70%. Fast and simple computation is also useful for industrial applications, resulting from dimension reduction of the spectroscopic data by PCA Furthermore, artificial neural networks showed high accuracy, close to 100%, after a few epoch calculations.
Musu et al. (Thu,) studied this question.