MODEL COMPARISON OF DEEP LEARNING FOR WATERFOWL CLASSIFICATION

Abstract

The purpose of this study is to compare the performance of image-based waterfowl classification models using deep learning. Specifically, we constructed a variety of waterfowl classification models using color images of waterfowl taken with a digital camera, in which deep learning with and without transfer learning were employed. Model performance of these models were compared with respect to accuracy. Model interpretability was examined using Grad-CAM. This analysis revealed that differences in model structure, datasets used for training, and image size affect the explanatory power and accuracy of the models.