AI System Enhances Brown Bear Monitoring and DNA Collection Efficiency

How Researchers Are Using AI to Track Brown Bear Behaviour and Enhance DNA Collection

Scientists have developed an artificial intelligence (AI)-based system to monitor brown bear behaviour and streamline DNA collection in remote regions. This innovative approach, detailed in a study by the Zoological Society of London, leverages AI models and camera-trap images to identify when bears stand on two legs near hair snares. Such behaviour often involves rubbing against trees, leaving behind fur samples crucial for DNA analysis.

Hair snares, which consist of barbed wire strands attached to trees, are a common non-invasive method for collecting bear fur. However, these setups require frequent site visits to retrieve samples and prevent contamination, a process that is both time-consuming and resource-intensive. In the Catalan Pyrenees, for instance, only 24% of visits to hair snare stations yield usable samples, underscoring the need for more efficient monitoring solutions.

The proposed AI system aims to address this challenge by analysing images captured by camera traps. Researchers trained the AI to distinguish between bears standing on two legs (bipedalism) and moving on all fours (quadrupedalism). Since no pre-existing dataset for bear poses was available, the team manually annotated 2,373 images collected between 2020 and 2023, marking 15 key anatomical points on each bear's body.

The system employs the YOLOv11 pose estimation model, which achieved a precision rate of 93.2% in detecting these key points. A separate multilayer perceptron (MLP) model then classified bear postures with 96.1% accuracy. This capability allows researchers to identify instances where bears likely interacted with tree-mounted hair snares, leaving fur samples suitable for DNA testing.

Designed for use in areas with limited or no cellular connectivity, the system transmits lightweight pose data and simplified visual representations via satellite links instead of full-resolution images. This reduces data transfer requirements from several megabytes to just about 100 bytes, making it more practical for remote applications.

By focusing field inspections on locations where bear activity is detected, the AI system could significantly reduce unnecessary site visits, saving time and resources. This efficiency boost is expected to enhance wildlife monitoring programmes.

The study concludes that AI-powered pose estimation offers a promising tool for ecological monitoring. Beyond automating parts of the DNA collection process, it provides valuable insights into bear behaviour that traditional hair snares alone cannot capture.