Contaminant detection in flexible polypropylene packaging waste using hyperspectral imaging and machine learning
Giuseppe Bonifazi1, Giuseppe Capobianco1, Paola Cucuzza1
1Department of Chemical Engineering, Materials & Environment (DICMA), Sapienza University of Rome, Via Eudossiana 18, 00184 Rome, Italy.
Waste Management (New York, N.Y.)
|February 11, 2025
Summary
This study uses hyperspectral imaging (HSI) and machine learning to automatically sort flexible plastic packaging (FPP) waste. The developed method effectively identifies polypropylene (PP) from contaminants, improving recycling efficiency and material recovery.
Area of Science:
- Materials Science
- Chemical Engineering
- Computer Science
Background:
- Flexible plastic packaging (FPP) is a major component of plastic waste, posing significant sorting and quality control challenges in recycling facilities.
- Accurate identification and separation of different plastic types within FPP waste streams are crucial for effective recycling and resource recovery.
Purpose of the Study:
- To develop and validate a hyperspectral imaging (HSI) based classification procedure combined with machine learning for automatic detection of contaminants within a polypropylene (PP) stream of FPP waste (FPPW).
- To assess the performance of the developed model in terms of classification accuracy, material recovery, and grade improvement for PP recycling.
Main Methods:
- Acquisition of hyperspectral images in the short-wave infrared (SWIR) range (1000-2500 nm) for representative FPPW samples containing PP and various contaminants.
- Preprocessing of spectral data using various algorithms, followed by Principal Component Analysis (PCA) for exploratory analysis.
- Application of a hierarchical classification model based on Partial Least Squares-Discriminant Analysis (Hi-PLS-DA) to differentiate PP from contaminants like polyethylene, polyester, aluminum, and multilayer films.
Main Results:
- The HSI-based Hi-PLS-DA model achieved a classification accuracy of 87.5%, correctly identifying 147 out of 168 flakes, verified by Fourier transform-infrared (FT-IR) spectroscopy.
- The model is predicted to achieve a PP recovery of 98.2% by weight with a grade of 94.4% by weight, a substantial improvement from the initial 77.2% grade.
- Minor classification errors were observed for filaments and multilayer flakes, indicating areas for potential model refinement.
Conclusions:
- Hyperspectral imaging combined with Hi-PLS-DA provides an effective and automated solution for sorting and quality control of flexible plastic packaging waste (FPPW).
- The developed method significantly enhances the purity and recovery rate of polypropylene (PP) in recycling streams, contributing to more sustainable plastic waste management.
- Further optimization may be needed to address classification challenges with specific FPPW morphologies like filaments and multilayer structures.
