Rapid classification of plastic bottles by laser-induced breakdown spectroscopy (LIBS) coupled with partial least squares discrimination analysis based on spectral windows (SW-PLS-DA)

Abstract

With the increasingly extensive use of plastic bottles, the recycling and reuse of plastics continue to raise additional concerns. The combination of laser-induced breakdown spectroscopy (LIBS) and chemometric methods has been applied to classify plastics. However, previous studies were prone to falling into over-fitting phenomena and did not focus on plastic bottles. In this regard, selecting suitable input variables by using spectral windows can reduce the influence of over-fitting phenomena on classification results because parts of plastic spectra that are analytically useless are eliminated. Conventional methods of selecting spectral windows can achieve better classification results; however, these methods suffer from complex processing schemes and poor automation. In contrast, selecting spectral windows based on the continuous wavelet transform (CWT), which can incorporate most information of spectral peaks and reduce the influence of noisy variables on the classification results, can resolve these problems. In addition, partial least squares discrimination analysis (PLS-DA), most widely used for spectral data, is a simple and stable multivariate analysis method. Thus, to reduce the influence of over-fitting phenomena on classification results, this paper presents an extension of PLS-DA, namely, SW-PLS-DA, which uses spectral windows based on the CWT to select input variables. For a comparison with other conventional methods of selecting spectral windows, the performance of introducing spectral windows into the PLS-DA model based on the CWT was evaluated. To validate the classification ability of SW-PLS-DA for plastic bottles, the SW-PLS-DA model was compared with the PLS-DA, support vector machine, and random forest classifiers. SW-PLS-DA has the highest accuracy (93.93%) and efficiency (0.0021 s), which indicates a good performance for the classification of plastic bottles.