Cooperative Fish and Wildlife Research Units Program: Headquarters
Education, Research and Technical Assistance for Managing Our Natural Resources


Zhang, Y., Y. Feng, S. Wang, Z. Tang, Z. Zhai, R. Viegut, E. Webb, A. Raedeke and Y. Shang. Deep Learning Models for Waterfowl Detection and Classification in Aerial Images. Information

Abstract

Waterfowl populations monitoring is essential for wetland conservation. Lately, deep 1 learning techniques have shown promising advancements in detecting waterfowl in aerial images. In 2 this paper, we present performance evaluation of several SOTA supervised and semi-supervised deep 3 learning models for waterfowl detection in aerial images using four new image datasets containing 4 197,642 annotations. The best-performing model, Faster R-CNN, achieved 95.38% accuracy in terms 5 of mAP. Semi-supervised learning models outperformed supervised models when the same amount 6 of labeled data were used for training. Additionally, we present performance evaluation of several 7 deep learning models on waterfowl classifications on aerial images using a new real-bird classification 8 dataset consisting of 6,986 examples and a new decoy classification dataset consisting of about 10,000 9 examples per category of 20 categories. The best model achieved accuracy of 91.58 % on the decoy 10 dataset and 82.88% on the real-bird dataset.