[M-IS17] Ocean Plastics, an earth science perspective

[MIS17-P03] Estimation of the abundance of plastic litter in the city using citizen science and deep learning

Ryunosuke Muroya¹, *Shinichiro Kako¹, Daisuke Matsuoka², Atsuhiko Isobe³ (1.Graduate School of Science and Engineering, Kagoshima University, 2.Japan Agency for Marine-Earth Science and Technology (JAMSTEC), 3.Research Institute for Applied Mechanics, Kyushu University)

Keywords:plastic litter in the city, deep learning, citizen scince, image processing

Approximately 80% of plastic litter entering the ocean is of terrestrial origin (Eunomia, 2016). Most of these are litter related to our daily activities that was mismanaged and discharged into the ocean via rivers (Lebreton et al., 2017). Currently, however, there is no method to accurately and objectively monitor and quantify the amount of plastic litter in terrestrial areas, in particular in the city where the source of marine plastic pollution. Existing methods for monitoring plastic litter include a quantification method for beach litter that combines drones and webcams with deep learning (Kako et al., 2022, Hidaka et al., 2022), but from the perspective of safety and privacy protection, it is difficult to apply these methods to monitoring litter on land, particularly litter in the streets. In this study, we attempted to establish a method to quantify the amount of litter in the city by combining images taken by ordinary citizens using a smartphone application and deep learning, with the aim of constructing a citizen-participatory litter monitoring network. Based on visual observations, an image analysis method capable of detecting six types of litter (can, plastic bottle, plastic bag, cigarette box, cigarette butt, and mask) in the city was constructed. The results of the accuracy validation through the visual observation showed that our image processing method based on deep learning is possible to detect and quantify 6 types of litter in the city with a high precision and recall rates of more than 70%.