Convolutional neural network applied to the classification of bird species

Abstract

The objective of this work is to design a classification system using a Convolutional Neural Network (CNN) applied to the classification of bird species. The classes of species used in this classifier are Ardea alba, Butorides striata and Dendrocygna viduata. The dataset used is composed of 6,000 samples of color images, being divided into two sets, one for training and the other for testing. CNN´s architecture consists of 5 layers of Convolutional and 5 layers of MaxPooling interspersed respectively, in addition to a Flatten layer and a Fully Connected layer. The results obtained by the successful classifier system can be visualized through the confusion matrix, for the three species. Likewise, the cross-validation performance measure for the classifier system corresponds to an average accuracy value of approximately 94% of the test images. It conclude that the classifier system behaved appropriately.