Cost-Efficient Detection of Plastics From Post-Consumer Packaging Waste Using Selected Bands in the Near-infrared Spectrum

Abstract

Current polymer identification methods for sorting and recycling purposes, such as near-infrared (NIR) spectroscopy and hyperspectral imaging, are data-intensive and costly. Spectral data from hyperspectral images and real waste samples were collected. In the first stage, characteristic wavelength regions for each polymer type were manually selected. Later, the selection process was automated. These regions were then used for model training and characterization of unknown spectra. Integrals of the raw spectra were computed for each wavelength region and further analyzed using the principal component analysis (PCA) method. The trained model is able to successfully allocate unknown spectra using either the k-nearest neighbors (kNN) algorithm or the convex hull method. The model was further optimized using a systematic parameter study. Results evidence that the model distinguishes between individual plastics and characterizes unknown spectra from real waste with up to 100% accuracy.