Nurdle Count - A machine learning approach to nurdle classification and quantification

The Nurdle Count project aims to develop a machine learning and artificial intelligence (AI) approach to classify and quantify nurdles. These small plastic pellets contribute to microplastic pollution in oceans and waterways. The project will use computer vision to detect nurdles, distinguish them from other particles, and provide an accurate count to users via the Nurdle Patrol Website and Mobile Applications. By automating the counting process, this project seeks to improve the efficiency and accuracy of nurdle data collection, ultimately supporting environmental research, advocacy, and policy change.

The goals and objective of this project include:

Data Collection: Researchers will collect nurdle images from various sources, including previously submitted images to the Nurdle Patrol Website and Application, in-house images of nurdles from Matagorda Bay, and pictures of nurdles provided by the NOAA Marine Debris Program. Researchers will develop a standard operating procedure (SOP) for capturing nurdle images, which will be used when communicating with citizen scientists using the final Nurdle Count product. The SOP will be distributed to the citizen scientists who volunteer to collect training data assisting the researchers in submitting quality nurdle images.

Image annotation: Researchers, subject matter experts, and citizen scientists will perform quality control and quality assurance of nurdle images, building a Nurdle Annotation Library to guide the labeling process and inform the machine learning and AI applications.

Model experimentation and training: Researchers will experiment with computer vision models and transfer learning models that have shown high accuracy in microplastic detection and quantification in previous methodologies. They will evaluate each model to see which detection model provides the highest accuracy.

Integration with Nurdle Patrol: The AI model will be integrated into the existing Nurdle Patrol Website and the Nurdle Patrol Mobile Applications, allowing users to upload images of their collected nurdles and receive an automatically logged accurate nurdle count for the image.

Link to Nurdle Patrol Website