Non-destructive discrimination of plastic wastes by combining near-infrared spectra measurement and neural network analysis

Abstract

At present, there is a strong social need to develop a rapid plastic- wastes discrimination system for recycling plastics. The possibility to discriminate many kinds of plastics rapidly by combining near-infrared spectra measurements and neural network analyses was examined. For that purpose, the near-infrared spectra in the 1.3-2.3 μm wavelength region were measured for about 300 samples of 51 kinds of plastic, and normalized second-derivative spectral data were trained in a three-layered perceptron-type neural network. As a result of a discrimination test using the spectral data averaged for sample groups produced by a cluster analysis, the system showed an overall performance of 77% to discriminate 51 kinds of plastic. The possibility to develop a practical plastics discrimination system using this approach was demonstrated.