Big Data, Tiny Targets: An Exploratory Study in Machine Learning-enhanced Detection of Microplastic from Filters

TL;DR

This work tackles automated detection and quantification of microplastics in SEM images within a filtration-based setup. It benchmarks YOLO-based object detectors (variants 8n, 9, 11, etc.) with two preprocessing schemes (binarization and histogram equalization) on a dataset of 152 images with 1,184 labeled particles, using an 80/20 train/test split and 250 training epochs. The study reveals model- and preprocessing-dependent trade-offs: 8n-seg typically yields strong overall accuracy ($mAP_{50}$, $mAP_{50-95}$) and F1, while 11n-seg can boost recall but may degrade segmentation quality, highlighting limitations such as data scarcity and image clutter. The authors conclude that dataset size/quality and cross-modal approaches are critical for generalization and outline future work toward multimodal data fusion and vision-transformer architectures to improve MP detection in SEM contexts.

Abstract

Microplastics (MPs) are ubiquitous pollutants with demonstrated potential to impact ecosystems and human health. Their microscopic size complicates detection, classification, and removal, especially in biological and environmental samples. While techniques like optical microscopy, Scanning Electron Microscopy (SEM), and Atomic Force Microscopy (AFM) provide a sound basis for detection, applying these approaches requires usually manual analysis and prevents efficient use in large screening studies. To this end, machine learning (ML) has emerged as a powerful tool in advancing microplastic detection. In this exploratory study, we investigate potential, limitations and future directions of advancing the detection and quantification of MP particles and fibres using a combination of SEM imaging and machine learning-based object detection. For simplicity, we focus on a filtration scenario where image backgrounds exhibit a symmetric and repetitive pattern. Our findings indicate differences in the quality of YOLO models for the given task and the relevance of optimizing preprocessing. At the same time, we identify open challenges, such as limited amounts of expert-labeled data necessary for reliable training of ML models.

Figures (6)

• Figure 1: SEM images of the 10 $\mu$m (left) and 1 $\mu$m (right) pore size Si filter membrane
• Figure 2: Images with differing illumination and contrast with (a) low illumination, (b) low contrast, (c) good contrast and illumination
• Figure 3: (a) original image, (b) binarization with the Otsu-method, (c) histogram equalization via CLAHE
• Figure 4: (a) segmentation mask predictions of model 8n, (b) overemphasizing of particles for the predictions of model 11n (dark blue region) of the same image.
• Figure 5: (a) model 8n: segmentation mask missing in the middle, (b) model 11n: the same image with the full segmentation mask