Improvement of recyclable plastic waste detection – A novel strategy for the construction of rigorous classifiers based on the hyperspectral images

Abstract The objective of our study was to evaluate the advantages of the proposed validation framework for a rigorous one-class classifier that operates on the large sets of multivariate pixels that are obtained using the hyperspectral imaging technique that are considered to be individual samples. The performance of the validation strategy was evaluated experimentally using the hyperspectral images of post-consumer waste polymers (high-density polyethylene and polypropylene), which have very similar physico-chemical properties. Rigorous classification models using the partial least squares approach were constructed for the training samples from modeled groups of polymer items. Their aim was to support the process of sorting polymer waste items and to potentially provide a framework to construct an intelligent laboratory system. The results that were obtained in this study provide evidence that the pixel-based approach improves classification in terms of sensitivity and specificity. The models that described high-density polyethylene polymer items and that were built to represent individual pixels were characterized by a sensitivity of more than 98.6% and a specificity of more than 99.5%, whereas the models that were constructed for the polypropylene polymer items had a sensitivity of more than 93.4% and a specificity of more than 99.9%.