A comparative study of deep learning-based network model and conventional method to assess beach debris...

Area of Science:

• Environmental Science
• Computer Science
• Marine Biology

Background:

• Conventional marine debris surveys are costly and inaccurate, relying on limited manual sampling.
• Existing methods struggle to provide comprehensive data on beach pollution levels.

Purpose of the Study:

• To develop an automated method for detecting and quantifying marine debris using deep learning.
• To compare the accuracy and efficiency of the automated method against conventional survey techniques.

Main Methods:

• A deep learning-based network model was developed for automatic detection of beach debris.
• The model was trained and validated through fieldwork in Korea.
• Performance was evaluated using mean average precision (mAP) and error rates compared to manual surveys.

Main Results:

• The developed network model achieved a precision of 0.87 (87%) mAP for item classification.
• The automated method demonstrated an error rate of less than 5% compared to actual surveys.
• This study presents the first Korean fieldwork comparing automatic and conventional marine debris assessment methods.

Conclusions:

• The deep learning model offers a cost-effective and accurate alternative to traditional marine debris surveys.
• The findings provide crucial data for developing effective beach debris management strategies and policies.
• Automated detection systems can significantly improve the monitoring of marine pollution.