Résumé:
Waste sorting is a crucial process for efficient waste management, with automation being
a significant factor for waste companies. Advancements in robotics, artificial intelligence, and
autonomous driving have enabled underwater waste cleaning robots to accurately locate and
recognize underwater waste. The image segmentation method, compared to deep-learning
based target detection, provides a more refined and accurate approach to target recognition,
improving environmental perception and efficiency in underwater waste sorting. To simplify
the process, we proposed two trained models (YOLOv8and Mask RCNN). We used them to
segment underwater waste and then facilitate the process of sorting the waste into different
types such as mask, glass bottles, plastic bottles, electronics, metal, tire, plastic bags and waste.
The proposed system is tested on Underwater_debris dataset. Our YOLOv8 model achieved
84% of average precision after 200 training epochs and an average precison that exceeds 83.3%
for our Mask RCNN model after 100 training epochs.
The
underwater waste separation
process
is
made
and more intelligent by our proposed waste segmentation approach.
