Eureka! Institutional Repository

In modern era, many bird species face the danger of extinction. The need of study and maintaining the biodiversity, contributed in development of automated observation systems. In current work the effort is done in order to classify 3 different bird species of Greece. The felicitous extraction of features require the correct processing of the data base. In this work, have been collected only specific parts of the recordings (trills) and with them the classification and recognition system is made. So for each recording firstly the noise level is reduced with the pre-emphasis filter where necessary and then using the Hilbert follower trills are obtained. Then, the fundamental frequency from each trill is computed via the autocorrelation. The features that are used are the average value of the fundamental frequency, the standard deviation, the bandwidth and the rapidness of changing the fundamental for each trill. Several tests were made with different neural networks (specifically, with different number of nodes) and the results are discussed. Also two different approaches in experiments were done (using different number of inputs and outputs of the neural network). The results are very encouraging for future use and improvement of the system