Identifying animal species in camera trap images using deep learning and citizen science

Willi and colleagues combined deep learning with citizen-science labeling to build and refine species-classification models across several camera-trap projects. They showed that crowdsourced annotations from platforms like Zooniverse can train accurate classifiers, and that integrating model predictions with human verification creates an efficient, scalable pipeline. The study examined how confidence thresholds let automated systems handle easy images while routing ambiguous ones to volunteers, optimizing the division of labor. It also explored transfer of models among projects with differing species and environments. By demonstrating a practical human-machine collaboration, the paper influenced how large camera-trap programs structure their workflows. It is a frequently cited reference at the intersection of citizen science and automated camera-trap image analysis.