Identification of Aras Birds with Convolutional Neural Networks
Seda BAYAT, Gültekin Işık · 2020 4th International Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT) · 2020
In this study, the bird species seen in Aras River Bird Sanctuary of Iğdır were identified using convolutional neural networks. Acoustic monitoring is performed to examine and analyze biological diversity. Passive acoustic recorders are used for this purpose. Generally, analyzes are performed on the raw sound recordings collected with these recorders. In this study, bird species were classified using processed sounds instead of raw sound recordings. Basically, two experiments were done with raw and processed data to see the difference. For this, mel-spectrograms were created from raw and processed sound recordings. The performance of the model trained on the 10 and 30-second mel-spectrogram obtained from the raw sounds was 77.05% and 88.33%, respectively. In the processed sound recordings, these accuracy rates increased to 81.47% and 92.08%, respectively. By processing the raw sounds, the performance of the model has been increased by approximately 4%.