Home > Press Releases > details

Press Releases

February 4,2022
JAMSTEC
KAGOSHIMA UNIVERSITY

Development of a new AI-based method for high-accuracy estimation of the coverage area of drifting debris from coastal photographs

1. Key Points

Succeeded in using AI-based image analysis to detect beached debris (artificial objects such as plastics, bottles, and cans; and natural objects such as driftwood and shrubs) from coastal photographs taken at ground level.
Confirmed its applicability in estimating the coverage area of beach litter along the coastline.
The developed method is highly versatile, and the AI learning dataset can be used globally.
Expected to lead to the automation of estimating the existing amount of plastic litter on coasts and its outflows to the open ocean.

2. Overview

Research Assistant Mitsuko Hidaka, Researcher Daisuke Matsuoka, and others at the Information Engineering Program, Research Institute for Value-Added-Information Generation, Japan Agency for Marine-Earth Science and Technology, in collaboration with Associate Professor Shin'ichiro Kako in Research and Education Assembly, Kagoshima University, developed a new method for estimating the coverage area of drifting debris using deep learning (*1). These achievements provided major clues toward the realization of a new coastal environment monitoring system that utilizes artificial intelligence (AI) technologies.

Drifting debris has a significant impact on the marine environment, and poses a major problem to the maintenance of fishing, tourism, and landscapes. To date, although fact-finding surveys of drifting debris along the coast have been conducted worldwide, a versatile and practical technology for quantifying the existing amount of debris has yet to be established.

Therefore, this study applied a deep learning-based image analysis technique called semantic segmentation (*2) to develop a method that detects drifting debris along the coast at a pixel level in photographs taken at ground level with digital cameras. The results indicated that this method can be applied to estimate debris coverage area, and that it has additional applications for coastal images other than the coastal area used for learning or aerial images that were taken using drones. The technology developed herein is expected to see practical application as a universal technology that estimates the existing amount of drifting debris along the coast using photographs taken with a digital camera as well as various coastal monitoring data such as aerial images.

These achievements were published online on February 1 in Marine Pollution Bulletin.

Title:Pixel-level image classification for detecting beach litter using a deep learning approach

Authors:Mitsuko Hidaka1*, Daisuke Matsuoka1*,#, Daisuke Sugiyama1, Koshiro Murakami1, Shin’ichiro Kako2

*Eaually contributed
#Corresponding author

Affiliation:
1. Research Institute for Value-Added-Information Generation(VAiG), Japan Agency for Marine-Earth Science and Technology(JAMSTEC)
2. Graduate School of Science and Engineering, Department of Ocean Civil Engineering, Kagoshima University

URL: https://doi.org/10.1016/j.marpolbul.2022.113371

Furthermore, the training dataset for the AI created in this study was published online in the marine science database catalog SEANOE, which can be downloaded and used free of charge for non-commercial and research purposes only.

Title:The BeachLitter Dataset v2022

Authors:Daisuke Sugiyama1, Mitsuko Hidaka1, Daisuke Matsuoka1#, Koshiro Murakami1, Shin’ichiro Kako2

#Corresponding author

Affiliation:
1. Research Institute for Value-Added-Information Generation(VAiG), Japan Agency for Marine-Earth Science and Technology(JAMSTEC)
2. Graduate School of Science and Engineering, Department of Ocean Civil Engineering, Kagoshima University

URL: https://doi.org/10.17882/85472

【Supplemental information】

*1
Deep learning
This is a machine learning method that utilizes a multi-layered neural network, and has led to major breakthroughs that overwhelm conventional knowledge in various fields such as image and speech recognition, language comprehension, and behavior recognition. In recent years, it has also been utilized in earth science fields for detecting signs of typhoons (reported on December 19, 2018) and determining seismic motion (reported on January 16, 2019).
*2
Semantic segmentation
This is an image recognition method that associates a label or category with every pixel in an image. In recent years, deep learning-based semantic segmentation has been utilized in the recognition of cars, roads, pedestrians, landscapes, etc., during autonomous driving; and the recognition of lesions, organs, etc., in medical image processing.

Contacts:

(For this study)
Daisuke Matsuoka, Researcher, and Daisuke Sugiyama, Research Technician,
Research Institute for Value-Added Information Generation(VAiG), Information Engineering Program(IEP), JAMSTEC
Shin'ichiro Kako, Associate Professor, Research Field in Engineering, Science and Engineering Area Graduate
School of Science and Engineering (Engineering) Department of Engineering Ocean Civil Engineering Program
(For press release)
Press Office, Marine Science and Technology Strategy Department, JAMSTEC
General Affairs Division, Graduate School of Science and Engineering
(Engineering),National University Corporation Kagoshima University
Inquiry Form