Detecting contaminants in post-consumer plastic packaging waste by a nir hyperspectral imaging-based cascade detection approach

Abstract

Recycling of post-consumer packaging wastes involves a complex chain of activ-ities, usually based on three main stages, that is: i) collection from households or recovery from Municipal solid waste (MSW), ii) sorting and, finally, iii) mechanical recycling. Among these activities contaminants detection and separation play a pre-eminent role. The utilization of a Near InfraRed (NIR) – HyperSpectral Imaging (HSI) based methods, along with chemometrics and machine learning techniques, can fulfill both the two previously mentioned goals. In this paper, the HSI-based sorting logics, to apply, to implement, to set up and to perform an automatic separation of paper, cardboard, plastics and multilayer packaging are investigated. The built PLS-DA-based cascading classification model allows to recognize polymeric fragments from cellulosic ones and to identify multi-layer materials (i.e. laminated plastic and laminated cardboard). The misclassified fragments are constituted by laminated plastics. The set up cascade model reached in prediction a Recognition and a Re-liability of 0.933. The proposed NIR-HSI-based approach can represent an optimal, reliable and low-cost answer to systematically identify impurities and composite materials inside plastic waste streams.