Human-in-the-loop eider duck counting in Arctic Canadawith an open-vocabulary multispecies wildlife detector

Accurate monitoring of eider duck populations in Arctic Canada is essentialfor understanding ecosystem health and supporting conservation efforts in arapidly changing climate. Traditional manual counting from aerial imagery istime-consuming, labor-intensive, and prone to observer bias. In this work, wepresent a human-in-the-loop wildlife counting system that integrates an open-vocabulary multi-species object detector to streamline and enhance the accuracyof eider duck surveys. The system leverages a pretrained open-vocabularymodel, enabling the identification of both target and incidental species withoutretraining, and employs human validation to correct and refine automated detec-tions. This collaborative workflow combines the scalability of machine learningwith expert ecological knowledge, reducing annotation effort while maintain-ing high accuracy. Field validation using aerial imagery from Arctic Canadademonstrates that our approach can significantly accelerate population assess-ments, improve consistency across surveys, and facilitate adaptive monitoring inremote environments.