A Comparative Study on Deep Learning Techniques for Bird Species Recognition

Naturally, birds appear around us at different locations in a variety of sizes, shapes, and colors. Bird species recognition provides crucial information on the state of the environment. Manual collection and processing of bird species data for the identification of birds is a huge task for ornithologists. Automatic bird recognition systems reduce their burden to some extent by collecting, processing bird related information and identifying the birds automatically. In this view, this paper presents a comparative study of the performances of MobileNet, AlexNet, InceptionResNet V2, Inception V3, and EfficientNet on bird species recognition. We gathered 11488 images of 200 bird species from the Kaggle dataset and increased the number of images to 40000 using data augmentation techniques. The experiment results shows that MobileNet and EfficientNet are the quickest training models. EfficientNet is outperforming the other models with test accuracy of 87.13%.