DuckNet: an open‐source deep learning tool for waterfowl species identification in UAV imagery

Abstract

Understanding how waterfowl respond to habitat restoration and management activities is crucial for evaluating and refining conservation delivery programs. However, site‐specific waterfowl monitoring is challenging, especially in heavily forested systems such as the Mississippi Alluvial Valley (MAV)—a primary wintering region for waterfowl in North America. We hypothesized that using uncrewed aerial vehicles (UAVs) coupled with deep learning‐based methods for object detection would provide an efficient and effective means for surveying non‐breeding waterfowl on difficult‐to‐access restored wetland sites. Accordingly, during the winters of 2021 and 2022, we surveyed wetland restoration easements in the MAV using a UAV equipped with a dual thermal‐RGB high‐resolution sensor to collect 2360 digital images of non‐breeding waterfowl. We then developed, optimized, and trained a RetinaNet object detection model with a ResNet‐50 backbone to locate and identify seven species of waterfowl drakes, waterfowl hens, and one species of waterbird in the UAV imagery. The final model achieved an average precision and average recall of 88.1% (class ranges from 68.8 to 99.6%) and 89.0% (class ranges from 70.0 to 100%), respectively, at an intersection‐over‐union of 0.5. This study successfully surveys non‐breeding waterfowl in structurally complex and difficult‐to‐access habitats using UAV and, furthermore, provides a functional, open‐source, deep learning‐based object detection framework (DuckNet) for automated detection of waterfowl in UAV imagery. DuckNet provides a user‐friendly interface for running inference on custom images using the model developed here and, additionally, allows users to fine‐tune the model on custom datasets to expand the number of species classes the model can detect. This framework provides managers with an efficient and cost‐effective means to count waterfowl on project sites, thereby improving their capacity to evaluate waterfowl response to wetland restoration efforts.