Automatic Identification of Bird Species using Audio/Video Processing
Nikitha Sharma, Aditi Vijayeendra, Vishnu Gopakumar, Prakhar Patni, Ashwini Bhat
Abstract
There are about 10,000 to 13,000 different species of birds in the world. Identification of bird species has been a taxing ordeal for ornithologists and domain experts for decades. Hence, automation of bird species classification will greatly help in enhancing ecological surveys. This paper presents a method to automatically identify bird species from a video recording of the bird by applying image and audio processing and classification techniques. The image and audio classification models, built using pre-trained neural networks - ResNet50V2 and EfficientNetB0, are trained and tested on an image and audio dataset containing 137 bird species. The datasets were curated using multiple data sources to expand the reach of the proposed model. The test accuracy rates of the two models were 97.1% and 92.4% respectively with a final overall model accuracy of 90%.