[M-IS30] Ocean Plastics, an earth science perspective

[MIS30-02] Estimation of marine debris volumes on beaches using unmanned aerial vehicles by utilizing image processing based on deep learning

*Kako Shinichiro¹, Shohei Morita¹, Tetsuya Taneda¹ (1.Graduate School of Science and Engineering, Kagoshima University)

Keywords:unmanned aerial vehicle, deep learning, image processing, plastic marine debris

Marine debris has emerged as a problem of global concern. To address this, a system that can accurately, rapidly, and objectively calculate the abundance of marine debris on the beach, determine the efficiency of beach surveys, and prioritize beaches that need to be cleaned, was developed. A three-dimensional model and an orthoscopic image of the beach, constructed by an aerial triangulation process using location data derived from an unmanned aerial vehicle (UAV) observations, enabled us to compute the volumes of marine debris by detecting the edges of marine debris through image processing based on deep learning. The accuracy of the method was verified by estimating the volumes of test debrisplaced on the beach in known quantities. Our system shows the potential of estimating marine debris volumes to within an error of 5% or less. It has significant positive environmental implications in terms of addressing the issue of marine debris.