Empowering citizen scientists: A web app for plastic litter classification

The accumulation of plastic pollution globally poses a threat to all natural ecosystems. Despite collective efforts to address the plastic pollution problem, large amounts of plastic waste collect and accumulate in various environments, eventually leading to oceans and coastlines. One key component of the plastic waste mitigation strategy lies in the ability to collect and report plastic waste data at the granular level. In the context of growing Artificial intelligence (AI), data-driven decision-making has become a primary source of inspiration for positive action by policymakers and community members. Current plastic waste categorizations are predominantly based on manual image collection and annotation. This manual process hinders efficient data acquisition and analysis, limiting the potential impact of AI-powered solutions. The main objective of this paper is to design and deploy an AI-enabled web application called Machine Learning for Mitigating Litter (ML²). The novelty of our approach is the ability to directly classify plastic waste categories from our images within the web application. This allows the data set to grow rapidly with less input from citizen scientists. Our machine learning model, which is built on approximately 8000 locally sourced (and annotated) images, can predict various categories of plastic waste with more than a 90% confidence rate. The web application enables shareable dataset generation of plastic waste from developing regions of the world using a citizen science approach. Furthermore, optimization techniques were investigated towards enhancing the ML model and visualization methods to promote rapid adoption of the web application in diverse plastic mitigation initiatives.