Truace
Subscribe

YOLO-based automated microplastic particle detection in microscopy

Source article: Development of a YOLO-guided automated (microplastic) particle analysis workflow

Microplastics (MPs) are an emerging pollutant of global concern, creating an urgent need for rapid and accurate monitoring workflows. Deep learning-based computer vision has demonstrated strong performance in finding particles in microscopy images, but its use as a front-end module for automated IR/Raman microscope-based MP analysis remains insufficiently developed, particularly in workflows that convert image-level particle detection results into microscope-executable operations for particle morphological chara…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1989· cite & verify
Development of a YOLO-guided automated (microplastic) particle analysis workflow
Bottled water under a microscope by NnnuUuaaAmMm. CC BY 4.0 · https://creativecommons.org/licenses/by/4.0
Trace impact readingContested
P 68The P score combines the specificity and measured human impact of the grounded problem claim with the strength of this Trace’s cited sources.

Both sides are scored from claims and sources, not community votes.

G 65The G score combines the specificity and measured human impact of the grounded gain claim with the strength of this Trace’s cited sources.

In brief

Researchers built a workflow that uses a YOLO11 instance segmentation model as the front end for infrared and Raman microscope analysis of microplastics. They trained it on 626 bright-field micrographs containing 7010 particles from reference plastics deposited on filters and glass slides, then tested it against two baseline approaches.

Faster, more reliable microplastic monitoring matters because these particles are an emerging pollutant of global concern. The results are strongest on clean or reference samples, though, and it remains unclear how well the model handles the cluttered backgrounds typical of real environmental or consumer samples.

Main points

  1. The dataset of 626 micrographs with 7010 particles was split 7:2:1 into training, validation, and test sets.
  2. On the reference test set at IoU 0.7, the model reached precision 0.86, recall 0.93, and F1 0.89.
  3. A tap water test set with a relatively clean background gave an F1 of 0.83.
  4. The workflow covers per-field autofocus, recovery of true microscope coordinates, particle descriptors, and programmatic microscope control.
  5. Code and annotated data are released openly for others to adopt and extend.

The gain

A YOLO11 instance segmentation model found reference microplastic particles in microscope images with an F1 of 0.89, beating a threshold method (0.25) and Mask R-CNN (0.84), and was built into an automated analysis workflow.

The problem

Particle detection performance dropped when image backgrounds were dirtier, as seen in commercial salt samples, which limits reliability on messy real samples.

The rundown

The study targets a gap: deep learning can find particles in microscope images, but it had not been well connected to automated IR/Raman microscope workflows that turn detections into executable microscope operations for particle characterization and spectral acquisition planning.

Against baselines, the YOLO11 model scored an F1 of 0.89 on the held-out test set, compared with 0.25 for a threshold-based method and 0.84 for Mask R-CNN. On tap water samples it achieved precision 0.78, recall 0.90, and F1 0.83.

Beyond detection, the authors addressed the technical challenges of end-to-end automation, including autofocus on each field and converting YOLO-style outputs into real microscope coordinates. They released the full implementation, code, and annotated data openly.

What this doesn’t fix

Detection performance decreased with dirtier image backgrounds, as shown on a commercial salt sample, so the workflow's accuracy is weaker on less clean real-world samples.

Sources

  1. Peer-reviewedAnalytical Methods2026-10-15

The debate