Eyal Halabi - Second Place
Advisor - Prof. Ilan Shimshoni
Department of Information Systems
Automatic Detecting and counting endangered colonial breeding birds using machine learning methods
Eyal Halabi1, Inbal Schekler1,2, Ilan Shimshoni1, Nir Sapir1
1Department of Evolutionary and Environmental Biology, University of Haifa, 2Department of Information Systems, University of Haifa
Over the years, the populations of many ground-nesting bird species have declined substantially worldwide, including the Little Tern and Common Tern species which are both endangered species in Israel. An artificial island in Atlit salt factory is the main breeding site for these two tern species in Israel with more than 85% of the population. Monitoring this site has traditionally been manual, posing challenges for consistent year-to-year comparisons and requiring significant effort. To address this, we developed an AI-based algorithm to automatically detect and count the number of individual terns and breeding birds from each species.
In this study, our aim is to create an artificial intelligence-based algorithm to automatically detect and count the number of individuals of each species and the number of breeding birds over the years. We used advanced methodological tools from the field of image processing and deep learning to detect and classify the terns by their species and whether they are breeding or not.
Using two online cameras located on the island, utilizing advanced image processing and deep learning techniques, we automatically scanned the island four times daily, with these scans serving as input for our algorithm. We employed object detection using YOLO V8 to identify terns and used spatial calibration methods to improve classification accuracy based on size differences between Little Terns and Common Terns. In addition, we identified breeding birds by tracking them on the same spot along the day as breeding birds incubate and therefore tend to stay at the same location. Our automated system successfully counted the number of terns and breeding birds, with results comparable to manual counts. The automated counting system demonstrates high reliability and provides essential data for the management and conservation of these threatened species. This advancement can significantly enhance monitoring processes for waterfowl globally, offering a robust tool for wildlife conservati Institute, Agricultural Research Organization (A.R.O.) – Volcani Institute Organizationon efforts