Near-Infrared Spectroscopy and Neural Networks for Resin Identification

Postconsumer plastics recycling constitutes a small fraction of public recycling, primarily because of the costs associated with collecting, sorting, and processing. As a result, the cost of manufacturing us. ing virgin material is often less than using recycled material. Plastic waste must be sorted to achieve the highest value recycled resin. Optical and artificial neural network technology may aid in reducing recycling costs by increasing the sorting speed and decreasing the fraction of im purity resins. A system for sorting waste plastics using nearinfrared (nearlR) re. flectance spectra and neural networks has been developed at Sandia National laboratories. in this article, optimization of a backpropagation neural network used for sorting the nearlR spectra of postconsumer plastics is presented. The number of hidden layers, transfer function type, and preprocessing techniques were varied. For classification of resin spectra using neural networks, the most important factor appears to be data preprocessing.