PlastOcean: Detecting Floating Marine Macro Litter (FMML) Using Deep Learning Models

Plastic Waste Management has been a challenging topic of discussion for the last decades. Today we can use state-of-the-art Deep Learning models to identify and quantify floating plastic ocean patches using satellite images. These techniques can help in making ocean cleanup efforts more targeted and efficient. Today, most institutions working on ocean cleanup, use Manta Trawls for quantifying and manually sampling the plastic garbage patches in the ocean waters. Due to their vastness and remoteness, manual quantification methods for maritime garbage patches are ineffective, making thorough assessments difficult, expensive, and very time-consuming. This approach removes the involvement of human labour in identifying the garbage patches. It also saves the overall cost of the cleanup because identifying patches can further help in optimizing the most efficient routes and strategies for cleanup operations. We trained a CNN Sequential model with 1026 web-scraped images of FMML to predict if the FMML is a plastic or a non-plastic. This study achieved an accuracy of 97.4%.