Quantitative Evaluation for the Peak Selectionin the Raman Spectroscopy Classification of Plastics Based on the Support Vector Machine

抄録

　To perform mechanical recycling of massive amounts of waste plastics, we developed an identifier oriented on industrial use and based on Raman spectroscopy. For the purpose of designing a practical classification technique that is accurate, fast, and robust against external noise, a simple comparison of spectroscopic peak intensities is considered to be more useful than a complicated multivariate analysis. However, such peak choosing is carried out empirically so far based on expert spectroscopic knowledges. In this report, we successfully demonstrate the classification of three kind plastics using only two peaks that are selected based on support vector machine (SVM) of machine learning; three types of plastic, polypropylene (PP), polystyrene (PS), and acrylonitrile‐butadiene‐styrene copolymer (ABS), are categorized well even in the case of the large artificially generated noise conditions. We could choose automatically a set of two peaks among six sets of four peaks that become candidates in the Raman spectra. Margin values in SVM allow obtaining the decisive information on how to select the peaks in the spectra quantitatively, for distinguishing different types of plastics.