Application of Vision Transformers to Contamination Detection in Densely Cluttered Waste Scenes

With the increasing global waste production, there is a rising need for improved waste management solutions to address this growing issue. In the United States, less than 35% of recyclable materials are actually recycled, leading to higher levels of pollution in both soil and oceans. This growing issue has serious environmental repercussions, exacerbating pollution in terrestrial and marine ecosystems. The root of this issue lies in the inefficiencies of the waste sorting procedure, which involves the challenging task of separating materials such as different kinds of plastic and polystyrene from the cluttered waste environment. One way to address this problem is by automatically detecting contamination in recyclable waste before it is compacted in the collection truck hopper. In this paper, we utilize state-of-the-art computer vision-based models to identify contamination within a densely cluttered waste environment. We conduct a comparative analysis with leading object detection models, demonstrating that Vision Transformers achieve superior accuracy.