AI methods have had a great deal of success in automating the arduous task of labelling biological and medical images. Camera trap images require a similar labelling task where many different labels need to be assigned, sometimes to a single image. This project will explore how an interactive system can be developed to assist in the semi-automatic labelling of multiple species. This will involve using state-of-the-art explainable AI (XAI) methods combined with multi-class active learning to assist in choosing labels or highlighting which images need human intervention. It will also use XAI methods to help explain why images were labelled in a particular way so that a better understanding of the strengths and weaknesses of AI methods can be determined.
Policy Impact of Research:
The main limiting factor in the use of camera-traps and other biodiversity sensors (e.g. acoustic sensors) for monitoring biodiversity is data classification. This project will combine machine learning tools and the use of citizen science sites – e.g. Zooniverse, and have wide applications to global monitoring networks.