Thrilled to showcase how artificial intelligence can support wildlife monitoring and conservation ! 🤗 🐧 🐢

🎉 Last day at AIMS for our student Matthew (co-supervised with Emmanuel Dufourq)! He presented his 2-year project: a lightweight pose estimation model to monitor African penguins 🐧 + custom Raspberry Pi camera system for remote footage capture 🎥
Amazing work — well done Matthew! 👏 #AI #Conservation

This study was part of Kukanya Zondo's research project for his Master's in Mathematical Sciences at AIMS South Africa
, where he graduated with distinction— congratulations! 💪

Thanks to Damien Chevallier and the CNRS team, 6 hawksbill turtles were equipped with onboard cameras, accelerometers, gyroscopes, and pressure sensors—enabling accelerometer signal validation. A first for this species!

Pre-training on data from green turtles or humans boosts accelerometer-based behavior identification for hawksbill turtles. Fine-tuning the model outperforms training solely on hawksbill data.

🚩 New paper published in JEB !
journals.biologists.com/jeb/article/...
We test transfer learning with deep learning to automatically identify the behaviors of endangered species like Hawksbill sea turtle using accelerometer data.

With minimal preprocessing, F1-score of 81.1% and Global accuracy of 97.2%. The 6 behavioral categories identified: Breathing, Feeding, Gliding, Resting, Scratching, and Swimming.

🚩 🚩Improving ecological knowledge of sea turtles
V-net : a new method based on deep learning to infer sea turtle behavior from multi-sensor loggers

www.sciencedirect.com/science/arti...