Project Overview
I led the design and development of a compact system that detects and counts microplastics specifically microfibers in water samples using a high-resolution camera, custom image processing, and a YOLO-based deep learning model. We enhanced an existing detection setup by enabling it to recognize complex fiber clusters and bundles. The goal: make pollution detection accessible for anyone, anywhere.
The image above shows our YOLO model detecting and labeling both microfibers and hair in a variety of sample images, with confidence scores displayed for each detection. Our system reliably distinguishes between microfibers (blue boxes/labels) and hair (cyan boxes/labels), even when they appear similar, demonstrating robust performance in real-world samples.
Detection Demo
Live detection of microfibers in a real-world water sample.
Our YOLO-based model achieved a 0.852 precision at full confidence across all classes, demonstrating high reliability in detecting microfibers and distinguishing them from similar particles like hair.
Results & Impact
- Consistently detects microfibers & fragments in live video with high accuracy (>90% mAP).
- Designed for easy retraining and open-source extension to new pollution types.
- Portable and low-cost—built for use in the field by researchers, educators, or citizen scientists.