Application of deep learning models to detect coastlines and shorelines

Area of Science:

• Earth and Environmental Sciences
• Computer Science
• Artificial Intelligence

Background:

• Accurate identification and monitoring of coastlines and shorelines are crucial for global coastal erosion assessment.
• Deep learning (DL) offers advanced capabilities for interpreting complex geospatial data.

Purpose of the Study:

• To propose indicators for coastline and shoreline identification.
• To develop DL models for automatic interpretation of coastlines and shorelines from high-resolution remote sensing images.
• To apply trained DL models for monitoring coastal erosion in Vietnam.

Main Methods:

• Training eight DL models (U-Net, U2-Net, U-Net3+, DexiNed) using high-resolution satellite images from Google Earth Pro.
• Utilizing object segmentation techniques for coastline and shoreline detection.
• Evaluating model performance based on accuracy and loss functions.

Main Results:

• The U-Net model, with an input image size of 512x512, achieved the highest performance (98% accuracy, 0.16 loss).
• The DL-interpreted results effectively supported coastline and shoreline identification for coastal erosion assessment in Vietnam over 20 years.
• Shorelines are suitable for observing tidal and wave changes, while coastlines are better for assessing erosion influenced by sea-level rise during storms.

Conclusions:

• The U-Net model demonstrates significant potential for predicting coastal changes.
• This study provides a framework for utilizing DL in coastal management and erosion monitoring globally.
• The findings highlight the distinct roles of coastlines and shorelines in understanding different aspects of coastal dynamics.