Avaliação de algoritmos de identificação de imagens para monitoramento de tubarões (AIMT).

Abstract

In natural aquatic environments, bathers still experience shark attacks quite frequently. Thus, there is a clear need for more effective monitoring of these animals in order to ensure protection and also promote greater safety for citizens. Thus, through Deep Learning, which permeates the foundations of Artificial Intelligence, it is possible to solve more complex tasks using artificial neural networks. In view of this, this work seeks to evaluate Machine Learning techniques and algorithms, more precisely Convolutional Neural Networks, in the process of monitoring sharks, compared to image identification. Nevertheless, it is configured as an applied research, of an exploratory nature and with a quantitative approach, in which the results obtained by the VGG 16, ResNet 50 and MobileNet V2 architectures are evaluated, during the classification process. Therefore, we obtained acceptable results from the applied analyses, such as the F1-Score, which reached values close to 0.90. Therefore, considering the performance of the models, we can see the potential of this experiment and its importance for the preservation process, which can be explored more comprehensively in related scenarios.