Using convolutional neural networks to recognition of dolphin images

Abstract

Classification of specific objects through Convolutional Neural Networks (CNN) has become an interesting research line in the area from information processing and machine learning, main idea is training a image dataset to perform the classifying a given pattern. In this work, a new dataset with 2504 images was introduced, the method used to train the networks was transfer learning to recognition of dolphin images. For this purpose, two models were used: Inception V3 and Inception ResNet V2 to train on TensorFlow platform with different images, corresponding to the four main classes: dolphin, dolphin_pod, open_sea, and seabirds. The paper ends with a critical discussion of the experimental results.