Avaliação do uso de visão computacional para detecção de filhotes de tartarugas marinhas.

Abstract

In Brazil, five out of the seven existing species of marine turtles can be found. Professionals specialized in this field carry out the counting of hatched offspring, cracked eggs, and stillbirths in nests. As a powerful tool, Artificial Intelligence can be used to assist in this process, enhancing the data collection for the safety of this species. This article presents a study using Deep Learning as a tool to aid environmental safety professionals in the counting of marine turtles. Specifically, four Convolutional Neural Network architectures (VGG16, ResNet, MobileNet, and DenseNet) were applied to assess their feasibility in addressing this problem. Satisfactory results were achieved in the experiments, reaching 0.97 in the F1-Score, a performance measure evaluating precision and recall, indicating a promising performance in detecting marine turtles. These findings encourage further research in this area, aiming to advance the application of Computer Vision in creating monitoring systems for environmental safety.