Identifying plastic materials in post-consumer food containers and packaging waste using terahertz spectroscopy and...

Area of Science:

• Materials Science
• Analytical Chemistry
• Data Science

Background:

• Accurate identification of post-consumer plastic waste is essential for improving recycling purity and value.
• Variations in plastic characteristics (shape, additives) create spectral inconsistencies, complicating identification.
• Existing methods struggle with diverse plastic waste streams.

Purpose of the Study:

• To develop a high-precision system for identifying transparent polyethylene terephthalate (PET), transparent polystyrene (PS), and black PS from waste.
• To combine near-infrared (NIR) and terahertz (THz) spectroscopies with machine learning (ML) for enhanced plastic identification.
• To utilize explainable AI (XAI) to understand the contributions of NIR and THz waves in plastic differentiation.

Main Methods:

• Integration of near-infrared (NIR) and terahertz (THz) spectroscopy techniques.
• Application of machine learning algorithms, specifically XGBoost and Bayesian optimization.
• Utilizing explainable AI (XAI) for spectral feature analysis and identification validation.

Main Results:

• Achieved a precision score exceeding 90% for identifying transparent PET, transparent PS, and black PS.
• Identified specific THz frequencies (0.140 THz for transparent PS, 0.075 THz for transparent PET) crucial for identification.
• NIR spectroscopy effectively distinguished black PS from transparent plastics.

Conclusions:

• The combined NIR and THz spectroscopy with ML offers a robust solution for plastic waste identification.
• THz spectroscopy's effectiveness is material-dependent, providing complementary insights to NIR.
• The developed technology advances high-precision identification systems and guides future research in THz spectroscopy for recycling.