Machine Learning-Enhanced Raman Spectroscopy for Microfiber Detection: From Model Development to Coastal...

Area of Science:

• Environmental Science
• Analytical Chemistry
• Materials Science

Background:

• Microfibers are the most common microplastics found across ecosystems.
• Existing microplastic identification methods are often time-consuming and complex.
• Distinct microfiber morphology aids in their detection and analysis.

Purpose of the Study:

• To develop a rapid detection method for microfibers using Raman spectroscopy and machine learning.
• To reduce spectral interferences and improve microfiber identification accuracy.
• To validate the method in real-world environmental samples for efficient pollution monitoring.

Main Methods:

• Created a Raman spectral dataset of 15 plastic fiber types.
• Utilized an autoencoder (AE) model for spectral reconstruction and noise reduction.
• Developed and evaluated four machine learning models: SVM, RF, CNN1D, and CNN2D.
• Applied the best-performing model (CNN1D) to environmental samples from the East China Sea.

Main Results:

• The autoencoder effectively reduced spectral interferences.
• CNN models achieved high accuracy, with CNN1D reaching 99.03% in lab conditions.
• The CNN1D method correctly identified 85.71% of microfibers in real-world samples, including PE and PET.
• Analysis of East China Sea water samples identified 775 microfibers, primarily cotton and polyester, with an average abundance of 2.92 ± 2.30 items/L.
• Spectral processing time was reduced to under 5 minutes.

Conclusions:

• Established a rapid and efficient framework for microfiber detection and classification.
• The method significantly accelerates environmental monitoring of microplastic pollution.
• Integration of morphology analysis offers potential for future source tracking systems.
• This approach contributes to developing effective microfiber pollution control strategies.