An annotated satellite imagery dataset for automated river barrier object detection

Area of Science:

• Environmental Science
• Remote Sensing
• Artificial Intelligence

Background:

• Millions of river barriers exist globally for flood control, hydropower, and irrigation.
• Lack of comprehensive data on barrier location and type hinders impact assessment.
• Small barriers like weirs are especially under-documented.

Purpose of the Study:

• To introduce the River Barrier Object Detection (RBOD) dataset for global river barrier identification.
• To facilitate the automatic detection of river barriers using satellite imagery and AI.
• To provide a benchmark for assessing the societal and environmental impacts of river barriers.

Main Methods:

• Developed a large-scale dataset (RBOD) with 4,872 high-resolution satellite images.
• Annotated 11,741 oriented bounding boxes across five river barrier classes.
• Validated dataset performance using five standard object detection algorithms.

Main Results:

• The RBOD dataset contains extensive, high-quality annotated data for river barrier detection.
• Object detection algorithms achieved benchmark performance on the RBOD dataset.
• Demonstrated the feasibility of using satellite imagery for global river barrier mapping.

Conclusions:

• RBOD is the first public dataset for river barrier object detection.
• This dataset is a valuable resource for understanding and managing river barriers worldwide.
• Enables improved assessment of river barrier impacts on ecosystems and human societies.