Visual WetlandBirds Dataset: Bird Species Identification and Behavior Recognition in Videos

Area of Science:

• Ornithology
• Computer Science
• Ecology

Background:

• Biodiversity loss necessitates effective animal monitoring.
• Existing video datasets lack detailed bird behavior annotations.
• There is a significant need for specialized bird video datasets.

Purpose of the Study:

• Introduce the first fine-grained video dataset for bird behavior detection and species classification.
• Facilitate the development of deep learning models for avian action recognition.
• Contribute to biodiversity preservation through enhanced monitoring tools.

Main Methods:

• Collected 178 videos in Spanish wetlands.
• Captured 13 distinct bird species.
• Annotated 7 specific bird behavior classes.

Main Results:

• Developed a novel dataset addressing a critical gap in avian research.
• Established baseline performance using state-of-the-art models for bird behavior recognition and species classification.
• Demonstrated the dataset's utility for deep learning applications.

Conclusions:

• The new dataset is essential for advancing AI-driven bird monitoring.
• This resource will support crucial decision-making for global biodiversity preservation.
• Future work can leverage this dataset for more sophisticated avian behavior analysis.