Shintaro Ogawa¹, Tan Chi Jie², Takumi Tomokawa², Sylvain Geiser², Sakmongkon Chumkamon², Ayumu Tominaga³,Eiji Hayashi^2
1Department of Creative Informatics, Kyushu Institute of Technology, 680-4, Kawazu, Iizuka-city, Fukuoka, 820-8502, Japan
²Department of Mechanical Information Science and Technology, Kyushu Institute of Technology, 680-4, Kawazu, Iizuka-city, Fukuoka, 820-8502, Japan
³Department of Creative Engineering Robotics and Mechatronics Course, National Institute of Technology Kitakyushu Colllege, 5-20-1 Shii, Kokuraminamiku, Kitakyushu-city, Fukuoka, 802-0985, Japan
pp. 238–241
ABSTRACT
This paper proposed a deep learning-based beach litter detector specifically designed for assessing litter levels on beaches effectively. This litter detector was created utilizing a HTC, also known as the Hybrid Task Cascade network, and its performance was compared to that of the traditional mask R-CNN network in order to judge its effectiveness. The findings uncovered that the HTC network possessed heightened sensitivity towards small and tiny litters taken within the RGB colored images.

ARTICLE INFO
Article History
Received 02 December 2022
Accepted 20 July 2023

Keywords
Mask R-CNN
HTC
Field robotics
Object detection
Deep learning
TACO

JAALR3408

Download article(PDF)