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

Area of Science:

• Environmental Science
• Computer Science
• Artificial Intelligence

Background:

• Beach litter poses a significant threat to marine ecosystems.
• Manual monitoring of beach litter is labor-intensive, time-consuming, and costly.
• Automated solutions are needed to efficiently assess and manage coastal pollution.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) technique for automatic beach litter identification.
• To enable pixel-wise classification of beach litter using semantic segmentation.
• To provide a scalable and efficient method for monitoring coastal pollution.

Main Methods:

• A deep learning model was trained for semantic segmentation using beach images.
• Eight segmentation classes were defined, including two specific beach litter categories.
• The model's performance was evaluated using Intersection over Union (IoU), precision, and recall metrics.

Main Results:

• The semantic segmentation model achieved high accuracy in identifying and classifying beach litter.
• Performance metrics (IoU, precision, recall) indicated effective segmentation capabilities.
• The method demonstrated robustness when applied to images from diverse locations.

Conclusions:

• AI-driven semantic segmentation offers a promising, efficient approach to monitoring beach litter.
• This technique can significantly reduce the time and cost associated with manual surveys.
• Further improvements can enhance the accuracy and applicability of AI in coastal environmental management.