Deep Learning meets Behavioural Ecology in the wild: methodological applications using the sociable weaver

ongoing

Studies of wild animals, from conservation to behaviour, are usually based on individually marked animals. This requires capturing, marking and sampling animals, which imposes limitations as these methods can be challenging, time consuming and impact individual welfare. Additionally, following and observing or video recording animals to obtain data is further constraining. Recent developments in artificial intelligence, in particular deep learning, have the potential do radically and rapidly change the way in which animals are studied in the wild. These new methods can push current boundaries by allowing not only less invasive methods of identification, but also obtaining large volumes of data and, importantly, collection of new types of data, allowing new questions to be addressed. In this proposal, we bring together a team of scientist and technical staff from three European countries and South Africa. Our aim is to develop highly innovative methods, based on rapidly advancing developments in deep learning, which can have a substantial impact on the study of wildlife biology. Specifically, we will streamline non-invasive methods (i.e. no capture) in order to obtain 1) individual re/identification in the field; 2) identification of individual attributes (e.g. sex, size); 3) automatic identification of behaviours (e.g. provisioning young, aggression). In addition 4) we will establish a pipeline to process large volumes of video data, combining individual and behavioural identification. The project is based on exchanges between staff with different expertise, and on work conducted both in the lab and in field. These exchanges are expected to boost creativity and result in meaningful skills transfer and a strengthened collaborative network. The expertise and the methods developed will have a meaningful and lasting impact in the field of behavioural and wildlife biology, contributing to increase Europe’s competitiveness and attractiveness.