Abstract

The lack of industrially-applicable, fast polymer classification systems is currently a major stumbling block in establishing both economically- and ecologically-useful waste recycling systems. With the advent of near infrared (NIR) spectral imaging for online classification, a method capable of distinguishing between different materials while simultaneously providing reliable size and shape information became available. In particular, polymer materials can be identified by their characteristic reflection spectra in the NIR without critical interferences from varying sample sizes and colours. A dedicated laboratory-scale prototype spectral imaging system has been developed and a number of classification algorithms have been evaluated for their applicability for polymer classification. Of the investigated algorithms, the Spectral Angle Mapper algorithm, supplemented by a threshold value and applied to the first derivatives of the normalised spectra, proved to be best suited for a rapid and reliable classification of polymers. Based on these achievements, an on-line system capable of classifying polymer parts delivered on a conveyor belt in real-time has been set up, which can be used, for example, as a sensor for fully-automated industrial polymer waste sorters.

© 2003 NIR Publications