Tracing Microplastic Aging Processes Using Multimodal Deep Learning: A Predictive Model for Enhanced Traceability

Area of Science:

• Environmental Science
• Materials Science
• Data Science

Background:

• Microplastic (MP) aging alters surface properties, impacting chemical release, contaminant adsorption, and environmental fate.
• Accurate tracing of MP aging is critical for assessing environmental risks but remains challenging.
• Physicochemical changes during MP aging are complex and influenced by various environmental factors.

Purpose of the Study:

• To develop a multimodal deep learning model for tracing aging factors of microplastics (MPs).
• To predict the major aging factors of aged MPs based on their physicochemical characteristics.
• To enhance the accuracy and reduce bias in assessing MP environmental behavior and risks.

Main Methods:

• Collected 1353 surface morphology images and 1353 Fourier transform infrared spectroscopy spectra from 130 aged MPs.
• Developed a multimodal deep learning model integrating image and spectral data.
• Validated the model's predictive accuracy for major aging factors.

Main Results:

• Physicochemical properties of aged MPs were shown to vary significantly with different aging processes.
• The multimodal deep learning model achieved 93% accuracy in predicting major aging factors.
• The multimodal model improved accuracy by 5-20% and reduced bias compared to single-modal approaches.
• Model predictions for naturally aged MPs aligned with known environmental aging conditions.

Conclusions:

• The developed multimodal deep learning model effectively traces microplastic aging factors.
• This approach offers improved accuracy and reduced bias in environmental risk assessments of aged MPs.
• Findings provide novel insights into understanding plastic aging processes and their environmental implications.