Innovative methods for microplastic characterization and detection: Deep learning supported by photoacoustic imaging...

Area of Science:

• Environmental Science
• Analytical Chemistry
• Computer Science

Background:

• Microplastic waste accumulation poses significant ecological challenges.
• Intricate microplastic morphology complicates rapid visual detection and classification.
• Existing methods struggle with limited datasets for microplastic characterization.

Purpose of the Study:

• To develop an advanced method for high-resolution microplastic imaging and classification.
• To overcome data limitations in microplastic detection using automated processing.
• To introduce a deep learning model for accurate identification of diverse microplastic forms.

Main Methods:

• Employing photoacoustic imaging for high-resolution microplastic sample visualization.
• Designing an automated data processing pipeline to generate sample masks and expand datasets.
• Proposing Vqdp2, a generative deep learning model with multi-task learning for microplastic prediction.

Main Results:

• Vqdp2 achieved excellent performance in microplastic classification accuracy.
• The model demonstrated superior feature extraction capabilities through multi-task training.
• Simultaneous constraint of model parameters in two training modes yielded outstanding morphological representations.

Conclusions:

• The developed Vqdp2 model effectively addresses challenges in microplastic detection and classification.
• Photoacoustic imaging combined with deep learning offers a promising approach for microplastic analysis.
• This research advances the visual characterization and identification of microplastic environmental pollutants.