/DuckNet

Deep learning-based tool for automated identification of duck species from drone imagery.

Primary LanguagePython

DuckNet

DuckNet is an open-source, deep learning-based tool for rapidly and accurately detecting, localizing, and classifying species of waterfowl in drone images—forked from the amazing work of BatNet. The baseline model (SSD with VGG16 backbone) is trained to identify seven waterfowl species and one species of other waterbird: American coot (Fulica americana), gadwall (Mareca strepera), green-winged teal (Anas carolinensis), northern pintail (Anas acuta), northern shoveler (Spatula clypeata), mallard (Anas platyrhynchos), redhead (Aythya americana), and ring-necked duck (Aythya collaris).

Citation: Loken, Z. J, Ringelman, K. M., Mini, A., James, D. & Mitchell, M. 2024. UAVs and Deep Neural Networks Effectively Detect and Identify Non-breeding Waterfowl. Remote Sensing in Ecology and Conservation (Submitted).

Corresponding author: Zack Loken, zack@savannainstitute.org
DuckNet development: Mike Mitchell, mmitchell@ducks.org

LICENSE for DuckNet: CC BY-NC-SA 4.0