Skip to content

AI-assisted Eider Counting Cut Annotation Time by 87.5% in Arctic Aerial Surveys

Large flock of common eiders flying along the coast
Research · Other research

An AI-assisted workflow for counting common eiders (Somateria mollissima) in aerial photographs reduced the time needed to annotate a full survey image by 87.5% compared with fully manual labeling, according to research first published in the AAAI conference proceedings on 14 March 2026. The system combines automated detections with human review so that experts correct the model rather than mark every bird from scratch.

The work used a collection of 750 high-resolution aerial images of eider aggregations from the Belcher Islands in Arctic Canada, taken in 2002, 2008 and 2021. Male and female eiders can be distinguished by plumage, but individuals may occupy only around 10 × 10 pixels in the photographs and can occur in dense groups, making manual demographic counts extremely time-consuming.

Human review remained part of the counting process

The researchers built the workflow around the OpenWildlife object-detection model and a human-in-the-loop system. Six fully annotated images, containing 8,339 individual ducks, were used to adapt the model. Users then reviewed model predictions in a modified LabelStudio interface, correcting local regions and feeding those corrections back into the system.

For evaluation, the team used 10 separate aerial images containing approximately 35,000 individually annotated eiders. The refined model itself reached a recall of 77.6%, a precision of about 66.0% and a counting error of 22.2% on this test set, which is why human validation remained part of the workflow. The 87.5% time reduction came from a small user study with five participants: average annotation time fell from 117.1 minutes per image with manual labeling to 14.7 minutes when participants started from model predictions and corrected them as needed.

The system has been deployed for eider monitoring

The workflow was developed in collaboration with the Arctic Eider Society and has been deployed for use in its survey work. The system can also distinguish males from females, providing demographic information in addition to total counts. The researchers designed the tools specifically for large aerial images and dense aggregations, with functions for image slicing, confidence filtering and regional corrections.

Sources and image credit

  1. Hsiao, J., Kalia, A., Zhang, Z., Sun, H., Patel, M., Clausi, D. A., Xu, L. L., Segal, B., & Heath, J. (2026). Human-in-the-loop eider duck counting in Arctic Canada with an open-vocabulary multi-species wildlife detector. Proceedings of the AAAI Conference on Artificial Intelligence, 40(47), 39978–39986. https://doi.org/10.1609/aaai.v40i47.41432Primary first publication of the study, published 14 March 2026. Reports the Arctic Canada aerial-image workflow, evaluation and deployment.
  2. Hsiao, J., Kalia, A., Zhang, Z., Sun, H., Patel, M., Clausi, D. A., Xu, L. L., Segal, B., & Heath, J. (2026). Human-in-the-loop eider duck counting in Arctic Canada with an open-vocabulary multispecies wildlife detector. AI Magazine, 47, e70077. https://doi.org/10.1002/aaai.70077Expanded open-access publication of the same underlying system and results, first published 20 August 2026.

Image: Kim Hansen (Slaunger) — migrating flock of common eiders (Somateria mollissima) at Sandhammaren, Scania, Sweden, photographed 29 March 2015; Wikimedia Commons; representative species image, not imagery from the Canadian aerial surveys · CC BY-SA 4.0 · Source

About this content: This story was produced with AI assistance within an editorial workflow developed by Wildlife Vagabond. Editorial responsibility remains with Wildlife Vagabond.How AI is used

Other research

SPECIES IN THIS STORY

Species in this story

Common Eider Somateria mollissima Explore species

Independent research and conservation news archive

Did you find this information useful?

Wildlife Vagabond is independently built and maintained. Voluntary support helps cover source verification, hosting and continued work on research and conservation news.

Support Wildlife Vagabond

The news archive will remain freely available.