An Open-Set Raman Spectroscopy Framework for Rapid Profiling of Microplastic Types and Aging States in Complex...

Area of Science:

• Environmental Science
• Analytical Chemistry
• Data Science

Background:

• Microplastics (MPs) are pervasive environmental pollutants with significant ecotoxicological impacts.
• Challenges exist in rapidly identifying MPs and assessing their aging within complex marine environments.
• MPs can accumulate in marine food webs and act as carriers for other contaminants.

Purpose of the Study:

• To establish a Raman spectral dataset for common polymer types and their aging stages.
• To develop and validate an AI-driven method for accurate MP identification and aging assessment in marine samples.
• To address the interference from complex environmental matrices during MP analysis in real seawater.

Main Methods:

• Creation of a Raman spectral dataset for various polymer types and aging stages.
• Training an attentional neural network (aNN) for polymer and aging state classification.
• Implementation of an open-set deep learning (OSDL) framework to handle complex marine matrices.
• Field-based validation using microplastic fragments from coastal seawater.

Main Results:

• The aNN achieved >97% accuracy for polymer identification and >93% for aging state identification under laboratory conditions.
• The OSDL framework demonstrated 94% accuracy for naturally aged MPs and 97% for nontarget particles in real seawater.
• Field validation yielded 90% accuracy for microplastic fragment identification.
• The OSDL approach outperformed conventional closed-set algorithms in complex samples.

Conclusions:

• Integrating Raman spectroscopy with the OSDL algorithm provides a robust method for marine MP identification and aging characterization.
• This approach offers a foundational tool for studying the environmental fate and impact of microplastics.
• Further research is needed to enhance the universality of the method for diverse polymer types and geographic locations.