Identification of Birds Using Spectrogram Image Processing and Artificial Neural Network Classifiers

Identifying birds in audio signals has been a challenge because the marking of certain species influences their classification. The proposed method consists of a mixture of sound processing for extracting useful signals from longer recordings with dynamic energy thresholds, image processing such as filtering and compression through image resizing for the treatment of the spectrogram and multilayer perceptron networks for the classification. The method shows encouraging results and flexibility, allowing the extension to a larger number of species, as well as applications to other fields involving sound or sound-like signals.