Application of hyperspectral imaging for identifying polymer blends in polypropylene flexible packaging waste for enhanced recycling

In this study, a fast and efficient analytical procedure based on hyperspectral imaging (HSI), combined with machine learning (ML), was applied to address the challenges in sensor-based sorting of flexible plastic packaging waste (FPPW). Among the challenges that recycling processes must overcome to obtain a homogeneous FPP as secondary raw material, one of the most critical is the detection and removal of contaminants, particularly when polymer blends are involved. The aim of the work was to effectively identify contaminants, i.e., polypropylene blends (Blend PP), within a homogeneous FPPW stream of polypropylene (PP). Hyperspectral images of plastic flakes, representative of a post-consumer product composed of FPP in PP contaminated by Blend PP, were acquired in the short-wave infrared range (SWIR: 1000-2500 nm). A chemometric approach was applied, starting from the application of different preprocessing algorithms to enhance the spectral characteristics of each class (i.e., PP and Blend PP classes), followed by principal component analysis (PCA) to explore the spectral data and the development of a ML classification model capable of distinguishing PP and Blend PP classes. The flake samples were also characterized by Fourier transform-infrared (FT-IR) spectroscopy in Attenuated Total Reflection (ATR), in order to evaluate the prediction performance obtained by HSI. The results demonstrated the effectiveness of the HSI-based strategy in supporting automated sorting and contaminant detection, as verified by FT-IR spectroscopy, significantly improving the recycling of FPPW composed of PP.