MicroScan AI: Artificial Intelligence for Microplastic Detection and Classification

Microplastic contamination presents a growing environmental challenge, while conventional identification and analysis can require specialized laboratory equipment, expertise, and substantial manual effort. MicroScan AI investigates how artificial intelligence, computer vision, and reproducible machine-learning methods can support the identification and classification of suspected microplastic particles from microscopy images.
The project currently includes a reproducible and testable image-classification software pipeline implemented in Python and PyTorch. The existing research infrastructure supports controlled training, validation and testing; dataset-integrity checks; configurable convolutional neural network (CNN) models; held-out evaluation; model checkpointing; image inference; automated testing; and documentation for independent validation.
The next phase will focus on scientific validation using authentic microscopy data. This includes establishing operational class definitions, developing and quality-controlling an annotated microscopy dataset, documenting image-acquisition and preprocessing protocols, training and comparing appropriate computer-vision models, and evaluating performance using predefined experimental criteria.
Particular attention will be given to reproducibility, dataset leakage, annotation quality, model generalizability, and independent validation. Depending on the characteristics and annotations of the resulting dataset, the research may subsequently investigate more advanced approaches such as object detection, segmentation, particle counting, and transfer learning.
Beyond environmental research, MicroScan AI will explore applications in STEM education and community-based environmental monitoring. The long-term objective is to determine whether an accessible AI-assisted microscopy workflow can support students, researchers, educators, and communities in investigating microplastic contamination while maintaining appropriate scientific and methodological safeguards.