Detecção em tempo real do coral sol (Tubastraea spp.) usando técnicas de visão computacional

Abstract

Sun coral is the popular name for corals of the genus Tubastraea, an invasive organism of concern due to its potential negative impact on coastal ecosystems. In order to prevent and control the damage caused by sun coral, monitoring activities require quick and effective measures. Computer vision techniques can be used as a solution, which has demonstrated potential in characterizing and monitoring organisms, habitats and ecosystems. Considering the need for early identification by bioinvasion, the main objective of the study was to develop a methodology for automatic detection of Tubastraea spp. in underwater images. Based on the YOLOv8 object detection model, training and validation were performed using a manually annotated dataset and subjected to augmentation, resulting in real and synthetic images of sun corals in the marine environment. The performance of the model was compared with a validation performed by an taxonomist on Tubastraea species, and the model was implemented in a graphical interface to facilitate its use. The results of the model are of high precision (>90%) for the detection of corals in the open and closed state. By comparing the performance of the model with human validation, it is concluded that it can be implemented to support the tasks of managers, taxonomists and professionals involved in monitoring bioinvasion by sun corals, in addition to being an alternative in cases of rapid evaluation of large quantities of images.