Fellows Seminar - Deep learning-driven identification and behavioral-physiological responses of Humboldt penguins to an intruder

©MNHN - F.-G. Grandin

By Yvon Le Maho and Víctor Planas-Bielsa, 2022 Fellows

Using the famous silhouette that appeared as either a hawk or a goose depending upon its direction of movement, Konrad Lorenz and Nicolas Tinbergen pioneered the behavioral response of birds to an intruder. Yet, although they received the Nobel Prize for this master study in ethology, they never agreed on the interpretation of the data. The first difficulty they faced is that they had to rely on the reading of rings or tags to investigate the individual responses of the birds to the intruder. Thus, they could not avoid the disturbance due to human presence. In addition, at that time, it was not yet possible to simultaneously monitor the physiological and behavioral responses of the birds, and thus to investigate whether, depending on their personality, they might behave differently with a similar physiological stress.

The challenge of our 2022 USIAS Fellowship was to combine our complementary expertise in machine learning (Víctor Planas-Bielsa) and in physiological ecology and bio-logging (Yvon Le Maho) to solve these issues. 

We present a new deep learning-based framework for the individual identification of animals using only images with higher accuracy and more robust than widely used algorithms in challenging conditions, such as those encountered in field studies. Additionally, our model provides a confidence score to assess whether an observed individual has previously been identified or represents a new individual, offering a valuable tool for wildlife monitoring and research. This is demonstrated through testing with small datasets involving two species that exhibit distinctive morphological features: Humboldt penguins and Hermann’s tortoises.

Concerning the physiological response to a threat, the only method that avoids human disturbance is monitoring heart rate. However, until recently, heart rate could only be monitored in wild animals using implanted loggers or external recorders with electrodes such as those used for joggers. In both cases, there was therefore no way to collect the data without any human disturbance. Thanks to a collaboration with the US Medtronic company, we were able to innovate by investigating the physiological response of Humboldt penguins to an approaching intruder using data-communicating implanted heart recorders.

Using remote-controlled rovers with different camouflages and ways of approach, we therefore show the behavioral and physiological responses of Humboldt penguins to an intruder without any bias due to human disturbance.

• More information about the two Fellows and their USIAS project: Remote data acquisition in penguin colonies.