Plastic classification via in-line hyperspectral camera analysis and unsupervised machine learning

An increase in the quality of recycled plastic is paramount to address the global plastic challenge and applicability of recycled plastics. A potent approach is mechanical plastic sorting but sufficient analytical techniques are needed. This study applies unsupervised machine learning on short wave infrared hyperspectral data to build a model for classification of plastics. The model can successfully distinguish between twelve plastics (PE, PP, PET, PS, PVC, PVDF, POM, PEEK, ABS, PMMA, PC, and PA12) and the utility is further proven by recognizing three unknown samples (PS, PMMA, PC). The experimental setup is constructed similar to an in-line industrial setup, and the machine learning is optimized for minimal data processing. This ensures the industrial relevance and is a stepping-stone to solve the global plastic challenge.

Automated summary

In this study, unsupervised machine learning was applied to classify plastics, successfully distinguishing between twelve types of plastics. The practicality of the model was further demonstrated by correctly identifying three unknown samples.