Advancing Microplastic Monitoring: Automatic Correction of the Aggregation and Discontinuity Issues Based on...

Area of Science:

• Environmental Science
• Analytical Chemistry
• Computer Vision

Background:

• Instrumental imaging is crucial for microplastic analysis but struggles with accurate segmentation of fibers and nonfibers due to particle aggregation and discontinuities.
• Existing methods face challenges in precisely quantifying microplastic characteristics when particles clump together or are fragmented.

Purpose of the Study:

• To develop an automated analytical method for characterizing environmental microplastics using instrumental imaging.
• To address and improve the detection accuracy of aggregated and discontinuous microplastic particles.

Main Methods:

• Developed and utilized a diffluent amodal instance segmentation former (DAISF) model trained on a large dataset of 130,536 manually labeled particles.
• Employed the Gauss-Laplace operator within the DAISF model to enhance segmentation performance, particularly for challenging particle shapes and arrangements.

Main Results:

• The DAISF model significantly improved the detection of aggregated fibers (71.8%) and nonfibers (89.2%), and discontinuous fibers (90.2%) and nonfibers (98.4%) compared to standard instrumental detection.
• Achieved superior recall and F1 scores, with low relative errors in particle number (≤19.1%), length (≤20.2%), and mass (≤12.4%).
• Demonstrated 31.0-, 3.1-, and 8.8-fold improvements over the instrumental approach for particle number, length, and mass quantification, respectively.

Conclusions:

• The developed computational method accurately characterizes environmental microplastics from instrumental imaging data.
• The DAISF model offers a robust solution for overcoming aggregation and discontinuity issues in microplastic analysis.
• This approach is valuable for efficient detection and rapid monitoring of microplastic pollution in the environment.