Manually Annotated Drone Imagery Dataset for Automatic Coastline Delineation.

Area of Science:

• Geospatial Science
• Remote Sensing
• Artificial Intelligence

Background:

• Accurate coastline delineation is vital for coastal management, disaster response, and ecological protection.
• A significant gap exists in manually annotated, high-resolution datasets for AI applications in coastal research.
• Existing datasets often lack the detail required for precise AI-driven shoreline analysis.

Purpose of the Study:

• To introduce the Manually Annotated DRone Imagery Dataset (MADRID), an open-source, high-resolution dataset for AI-based coastline delineation.
• To address the scarcity of annotated data for training AI models in coastal zone analysis.
• To facilitate advancements in coastal protection and marine ecosystem monitoring.

Main Methods:

• Captured high-resolution RGB imagery using Unmanned Aerial Vehicles (UAVs) across two distinct Polish coastal types (cliff and dune).
• Developed a novel polyline annotation technique for precise manual coastline annotation.
• Organized 3691 images into training and testing subsets, adhering to FAIR data principles.

Main Results:

• Successfully created and released the MADRID dataset, featuring 3691 manually annotated, high-resolution images.
• The dataset is pre-split for semantic segmentation tasks, enabling direct use in AI model training.
• Annotations utilize a novel polyline method for enhanced coastline representation.

Conclusions:

• The MADRID dataset provides a valuable resource for advancing AI in coastal research and applications.
• This dataset will support improved coastal protection strategies and disaster management planning.
• The findings offer crucial insights into the coastal dynamics of the Southern Baltic Sea region.