Tracking Variations in Pelican Horn Size Across the Breeding Season Using Deep Learning

Pelicans develop distinctive beak protrusions, or horns, during the breeding season, but their growth and regression patterns remain largely unquantified. This study leverages deep learning to track seasonal variations in pelican horn size at the Great Salt Lake. Using a dataset of approximately 5,000 images, we trained a YOLO-based neural network on a subset of annotated images to detect pelicans and segment their beak and horn regions. We quantified horn size by calculating the ratio of the horn’s area to the beak’s area, providing a standardized measurement unaffected by image scale. Our results reveal a clear seasonal trend in horn size, with notable growth early in the breeding season followed by regression. While the precise function of these horns remains uncertain—given that they are neither sexually dimorphic nor linked to aggression—our findings offer new insights into pelican mating and migration behaviors. This study demonstrates the potential of deep learning for noninvasive wildlife monitoring and morphological analysis.