Fig 1.
Schematic overview of the experimental and data analysis workflow for plastic waste characterization.
The process starts with the collection of complex municipal solid waste (bottom left) and the preparation of representative plastic samples (middle left). Attenuated Total Reflectance-Fourier Transform Infrared (ATR-FTIR) spectroscopy is used for rapid Spectrum Acquisition, generating both raw overlaid spectra and distinct standard spectra for ten polymer types. The acquired data then undergoes Preprocessing (e.g., Savitzky-Golay smoothing and First Derivative) to remove noise and baseline effects. Subsequently, Principal Component Analysis (PCA) is applied to the preprocessed spectra to extract key features and visualize class separation, as shown in the score plots for different preprocessing methods (a-d). This optimized data pipeline ultimately leads to the high-accuracy Classification of the ten target plastics (right).
Fig 2.
Representative photographs of municipal solid waste plastics collected from recycling stations in the Beijing–Tianjin–Hebei region.
(a) Mixed waste plastics in original form; (b) selected intact items and fragments before cleaning; (c) various packaging materials and plastic products included in the sample set.
Fig 3.
Examples of oil-contaminated plastic samples used for ATR-FTIR measurement.
Ten common plastic types: ABS, PA, PC, PE, PET, PP, PS, PU, PVC, and PTFE, after standardized cleaning and then surface-coated with 0.5 mL of corn oil to simulate real-world food oil contamination.
Fig 4.
Original ATR-FTIR spectra (4000−500 cm-1) of various types and numbers of actual plastic samples.
(a) 66 PET samples; (b) 68 PP samples; (c) 50 PTFE samples; (d)50 PU samples; (e) 50 PVC samples; (f) 50 PE samples; (g) 50 PS samples (h)41 ABS samples; (i) 50 PA samples; (j)50 PC samples.
Fig 5.
ATR-FTIR spectra of ten standard plastic samples in the range of 4000−500 cm−1.
From top to bottom: PU (polyurethane), PET (polyethylene terephthalate), PVC (polyvinyl chloride), PE (polyethylene), PTFE (polytetrafluoroethylene), PC (polycarbonate), PS (polystyrene), PA (polyamide), PP (polypropylene), and ABS (acrylonitrile butadiene styrene).
Fig 6.
ATR-FTIR spectra of corn oil (black spectrum) and olive oil (red spectrum) in the range of 4000−500 cm−1.
Major characteristic peaks are labeled in the figure. ATR-FTIR spectra of corn oil (black) and olive oil (red) in the wavenumber range of 4000–500 cm ⁻ ¹. Characteristic absorption peaks are labeled, corresponding to key functional groups: the C–H stretching vibrations at 2922 cm ⁻ ¹ and 2852 cm ⁻ ¹, the ester carbonyl (C = O) stretch at 1743 cm ⁻ ¹, the C–H bending vibration at 1458 cm ⁻ ¹, the C–O stretching mode at 1161 cm ⁻ ¹, and the CH₂ rocking vibration at 721 cm ⁻ ¹. The spectral similarities reflect the common triglyceride composition, while subtle differences may arise from variations in fatty acid profiles.
Table 1.
Performance of different single pre-processing method.
Fig 7.
Overlaid ATR-FTIR spectra (4000−500 cm−1) of all 525 plastic samples (covering 10 types) after S-G 15-point smoothing pretreatment.
Overlaid ATR-FTIR spectra (4000–500 cm−1) of all 525 plastic samples representing 10 different polymer types, after Savitzky-Golay (S-G) 15-point smoothing pretreatment. The spectra exhibit characteristic absorption bands associated with various functional groups, including C–H stretching (~2900–3000 cm−1), C = O stretching (~1700–1800 cm−1 for polyesters and polyamides), C–O stretching (~1100–1200 cm−1), and aromatic ring vibrations (~1600–1500 cm−1). The overlapping nature of the spectra highlights both common features among polymers and subtle spectral differences that can be exploited for classification using chemometric methods.
Fig 8.
Principal Component Analysis (PCA) score plots of ATR-FTIR spectra for ten plastic types after different preprocessing methods.
(a) S-G 11-point smoothing + PCA; (b) S-G 15-point smoothing + PCA; (c) First Derivative (FD) + PCA; (d) Standard Normal Variate (SNV) + PCA. Ellipses represent 95% confidence intervals for each class. The legend on the right of (b) applies to all subplots: ABS (black), PA (red), PC (green), PE (blue), PET (cyan), PP (magenta), PS (yellow), PTFE (dark yellow/olive), PU (purple), PVC (navy blue).
Table 2.
Prediction performance of RF models based on PCA for plastic type classification.
Fig 9.
Confusion matrix for the classification of ten plastic types on the test set using the S-G 15-point smoothing + PCA + RF model.
Rows represent the true class and columns represent the predicted class. Numbers 1-10 correspond to: 1-ABS, 2-PA, 3-PC, 4-PE, 5-PET, 6-PP, 7-PS, 8-PTFE, 9-PU, 10-PVC. Overall accuracy was 96.2%.
Fig 10.
Results of Independent Component Analysis (ICA) applied to the ATR-FTIR spectrum of a PVC sample contaminated with corn oil.
(a) First independent component (IC1), primarily representing the PVC spectrum; (b) Second independent component (IC2), primarily representing the corn oil spectrum; (c) ATR-FTIR spectrum of pure PVC; (d) ATR-FTIR spectrum of pure corn oil.
Table 3.
Prediction performance of BO-RF models based on ICA-extracted features for oil-contaminated plastic type classification.
Fig 11.
Performance of the S-G 15-point + ICA + PCA + BO-RF model for the direct identification of oil-contaminated plastics on the test set.
(a) Comparison of predicted class versus true class for test samples, with an overall accuracy of 92.5%; (b) Confusion matrix for the classification of ten oil-contaminated plastic types. Rows represent the true class and columns represent the predicted class. Numbers 1-10 correspond to: 1-ABS, 2-PA, 3-PC, 4-PE, 5-PET, 6-PP, 7-PS, 8-PTFE, 9-PU, 10-PVC.