Hybrid deep learning framework for environmental microplastic classification: Integrating CNN-based spectral feature...

Area of Science:

• Environmental Science
• Analytical Chemistry
• Data Science

Background:

• Accurate microplastic (MP) classification from FTIR spectra is hindered by spectral variability and limitations of current models.
• Existing methods struggle to capture both local and global spectral features effectively.

Purpose of the Study:

• To develop a robust and accurate method for classifying environmental microplastics (MPs) using Fourier-transform infrared (FTIR) spectroscopy.
• To address the challenges of spectral variability and environmental weathering in MP identification.

Main Methods:

• A hybrid deep learning framework integrating Convolutional Neural Networks (CNNs) and Transformer models was developed.
• The CNN component extracts local spectral features, while the Transformer module captures long-range dependencies.
• A diverse spectral dataset of 17 polymer types from various environmental matrices was constructed.

Main Results:

• The proposed CNN-Transformer model achieved 95.77% classification accuracy on the validation set.
• The model outperformed traditional machine learning methods in MP classification.
• Robustness was confirmed through 50 independent trials, showing stable and reproducible performance.

Conclusions:

• The hybrid CNN-Transformer architecture offers a reliable and scalable solution for rapid MP identification in diverse environmental contexts.
• This approach enhances feature extraction, improves classification robustness, and ensures generalizability.
• The findings have implications for pollution monitoring and regulatory assessment of microplastic contamination.