Application of mid-infrared spectroscopy in plastic waste discrimination: chemometric methods and exploration of their potential in hyperspectral applications

• Published in: Measurement : journal of the International Measurement Confederation Vol. 257; p. 118657
• Main Authors: Yang, Jian, Xu, Yu-Peng, Chu, Xiao-Li
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 15.01.2026
• Subjects: Chemometrics; Discrete wavelet transform; Mid-infrared spectroscopy; Plastic waste discrimination; Plastic waste recycling; Plastic waste discrimination; Discrete wavelet transform; Plastic waste recycling; Chemometrics; Mid-infrared spectroscopy
• ISSN: 0263-2241
• DOI: 10.1016/j.measurement.2025.118657

•Comparison of nine chemometric models for accurate plastic waste discrimination.•A shallow convolutional neural network achieves 100 % discrimination accuracy.•Discrete wavelet transform improves model robustness in plastic waste discrimination.•Findings support mid-infrared hyperspectral imaging in plastic waste discrimination.•Key spectral features revealed using model interpretability analysis methods. Plastic waste classification is crucial for recycling. The integration of spectroscopy and chemometric-based classification methods has garnered attention due to their non-destructive, environmentally friendly, and highly efficient analysis. Although near-infrared (NIR) spectroscopy technology for high-speed online analysis has been well-developed, it has limitations in identifying dark-colored plastics. In contrast, mid-infrared (MIR) spectrometers have an edge in this regard. Mid-infrared hyperspectral imaging (MIR-HSI) is increasingly used in online and portable applications, but it faces challenges in terms of wavelength range and resolution, which affects the spectral quality of on-site analysis. This study established a MIR spectral database containing 698 plastic samples from 4 sources and 9 types, and the data diversity was expanded through simulated noise and variational autoencoder (VAE). Nine chemometric models were assessed, and innovations were made in the algorithms. The combination of discrete wavelet transform (DWT) and partial least squares discriminant analysis (DWT-PLS-DA) exhibits optimal performance in the mid-infrared spectral region. Deep (DWT-DCNN1D) and shallow (DWT-SCNN1D) one-dimensional convolutional neural network models based on DWT achieve a discrimination accuracy exceeding 99 % for nine types of plastics in the long-wave infrared spectral band. Furthermore, the DWT-PLS-DA and DWT-SCNN1D models attain a perfect discrimination accuracy of 100 % for three plastic types (PE, PA, and PP) in the mid-wave infrared spectral band. The model interpretation provided valuable insights for feature selection.