Real-Time Behaviour Recognition on Bio-Loggers Enables Autonomous Audio Playback Experiments in Free-Ranging...

Area of Science:

• Animal behavior and ecology
• Bio-logging technology
• Machine learning applications in wildlife research

Background:

• Traditional audio playback experiments are limited by location and observation constraints.
• Previous methods required direct human observation or fixed automated devices.
• There was a need for a system that could conduct experiments in natural, unconstrained environments.

Purpose of the Study:

• To design and validate an autonomous audio playback system integrated into bio-loggers.
• To enable real-time behavioral state estimation for triggering experiments.
• To overcome the spatial and temporal limitations of conventional playback studies.

Main Methods:

• Developed an autonomous bio-logger with an on-board machine learning model for real-time behavioral state estimation (using accelerometers).
• Integrated high-cost sensors (video camera) and audio playback triggered by detected target behaviors.
• Conducted field experiments with black-tailed gulls (Larus crassirostris) in Japan, collecting multi-modal data (acceleration, GPS, video).

Main Results:

• The real-time behavior recognition system achieved high accuracy (macro F1-score = 0.91).
• Autonomous audio playback experiments were successfully conducted on freely ranging gulls (46 events).
• Analysis revealed clear behavioral responses to both predator calls and noise stimuli.

Conclusions:

• The autonomous bio-logger audio playback system represents a successful proof of concept.
• This technology removes location constraints, expanding possibilities for ecological and behavioral research.
• Future extensions could include other sensor modalities for more sophisticated in-situ experiments.