Plastic sorting is an essential part of waste management. However, the presence of multilayered plastic makes sorting much more challenging.
Multilayered plastic is widely employed to enhance the functional properties of packaging, i.e. combining functions in thickness, mechanical strength, and heat tolerance. Yet, materials containing multiple polymer species and need to be pre-treated before they can be recycled as mono-materials. Automated sorting of plastic packaging has significantly improved material sorting due to technological improvements, especially in Near Infrared-Hyperspectral Imaging (NIR-HSI) for material characterization. HSI is fast without sample preparation, which makes it highly valuable for the non-destructive identification of monolayer plastics; identification of multilayer materials however requires novel approaches for chemical pattern recognition.
NMF is a widely used method for the resolution of hyperspectral images. identify mono/multilayer plastics with HSIs; it extracts characteristic information from the recorded HSIs with the help of chemically relevant model constraints; it also allows chemically-driven data preprocessing to remove artifacts and enhance the chemical information within the HSI.
It however suffers from collinearity among concentration contribution maps. MB-NMF will suffer less from such collinearity, because, the correspondence among the different chemical species in the multiple data blocks can be used as a constraint for the presence or absence of particular polymer species. Using this constraint reduces the ambiguity in their resolution and will thereby considerably improve identification accuracy compared to NMF and single data block.
Finally, an Ftest will perform to distinguish between monolayer and multilayer objects. A number of plastics with know composition (mono/multilayer) were analyzed by MB-NMF and the results are discussed.
Keywords: Non-Negative Matrix Factorization, Selectivity, Curve resolution, Plastic sorting, Multilayer plastic, Multi-block data sets, Hyperspectral Imaging