An improved classification method of waste smartphone plastics based on near-infrared spectroscopy

The near-infrared spectra technique is an effective and non-destructive analysis method for polymer identification, and it has been applied in waste plastics classification. Since the near-infrared spectra are easily affected by instrumental noise, baseline drift and light scattering, spectral data preprocessing is indispensable for classification tasks. In this paper, a classification method is proposed to improve the accuracy of waste smartphone plastics classification. Savitzky–Golay smoothing, moving average smoothing, first-derivative, second-derivative, standard normal variate, and multiplicative scattering correction were used for noise reduction, baseline correction, and scattering correction. These algorithms were combined into five strategies for the spectral data preprocessing, and the optimal spectral data preprocessing strategy was selected based on the accuracy of mixed plastics classification. Successive projections algorithm and competitive adaptive reweighted sampling were used to extract the spectral feature and also the optimal spectral feature extraction algorithm was selected according to the accuracy of mixed plastics classification. The NIR spectra data of four waste smartphone plastics: polyamide, polycarbonate, acrylonitrile butadiene styrene, and polycarbonate/acrylonitrile butadiene styrene blend were analyzed to illustrate the performance of the proposed method. The results show that the accuracy of the proposed method is improved by 11.2% on average compared with the method without optimization.