Identificação de espécies de aves através do aprendizado de múltiplas instancias em redes neurais profundas

Abstract

The application of deep learning to image classification for the automated identification of bird species represents a field of artificial intelligence that employs advanced image processing and machine learning techniques, aiming at the automatic recognition of species through images. This work presents a model for image classification that combines convolutional neural networks, multiple instance learning and transfer learning for the automatic recognition of the diverse bird species found in Brazil. The results obtained showed an accuracy of 64,20% in the species classification capacity. This study emphasizes the importance of offering society a tool that can facilitate the identification of one of the most extensive biodiversity found on the planet.