Updating "BePLi Dataset v1: Beach Plastic Litter Dataset version 1, for instance segmentation of beach plastic...

Area of Science:

• Environmental Science
• Computer Vision
• Marine Biology

Background:

• Beaches accumulate significant plastic litter, which fragments into microplastics, posing environmental challenges.
• Effective monitoring of macroplastic litter on beaches is essential to prevent further degradation and environmental harm.
• Automated, objective image-processing methods are needed for detailed litter distribution analysis using remote sensing data.

Purpose of the Study:

• To introduce the BePLi Dataset v2, an updated resource for training deep learning models for beach plastic litter detection.
• To provide a comprehensive dataset with detailed annotations for developing automated image analysis techniques.
• To facilitate research in beach litter monitoring and management.

Main Methods:

• The BePLi Dataset v2 contains 3722 original images from the Northwest Japan coast.
• It includes 118,572 manual annotations for plastic litter at both pixel and individual object levels.
• Objects are categorized into thirteen distinct plastic classes, including "pet_bottle," "plastic_bag," and "fishing_net."

Main Results:

• The dataset supports the development of instance segmentation and object detection models for macro beach plastic litter.
• Annotations include bounding boxes and pixel-level masks, enabling various analysis levels.
• This resource aids in counting objects and estimating litter coverage on beaches.

Conclusions:

• The BePLi Dataset v2 is a valuable resource for advancing automated beach plastic litter detection.
• It enables the development of sophisticated deep learning models for environmental monitoring.
• The dataset supports diverse applications in marine pollution research and management.