Unraveling the secrets of deep-sea benthos through artificial intelligence

• Grant number: 23/09821-9
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: October 01, 2023
• End date: December 31, 2024
• Field of knowledge: Physical Sciences and Mathematics - Oceanography - Biological Oceanography
• Principal Investigator: Paulo Yukio Gomes Sumida
• Grantee: Raphaela Neves Lopes
• Host Institution: Instituto Oceanográfico (IO). Universidade de São Paulo (USP). São Paulo , SP, Brazil

Abstract

The deep ocean is the largest biome on the planet. Biodiversity hotspots found there, such as deep-sea corals, form essential three-dimensional habitats present in various parts of the globe, providing ecological niches for numerous species. Although the existence of these coral environments has been known for a long time, studies have been limited due to the difficulties of sampling in the deep ocean and the potential impacts on these extremely vulnerable and sensitive ecosystems. On the Brazilian coast, few studies have been conducted on the distribution of these corals and their relationship with environmental variables. However, technological advances have made it possible to increasingly analyze images collected by submersible vehicles in the deep ocean. These operations, often carried out by the oil and gas industry, require significant effort to identify and annotate organisms due to the high volumes of data generated. Thus, there is a growing need to develop methods to automate these analyses. The current project aims to identify the organisms present on two deep-sea coral banks in the Campos Basin using ROV videos provided by Petrobras and to implement artificial intelligence for the automated detection of these organisms. Initially, identification will be performed through georeferenced manual annotations using the software Annotation 1.09. These data will be integrated with environmental information from the banks to evaluate the influence of these factors on the distribution of the organisms. Artificial intelligence will be implemented through a detection algorithm based on YOLOv5, which will be trained and refined to detect organisms present in the videos. Subsequently, statistical tests will be conducted to evaluate the algorithm's performance, comparing its results with the manual annotations. In addition to providing valuable data about an ecosystem still relatively unknown in Brazil, this project is a relevant initiative for optimizing the use of data from these environments, which are particularly vulnerable to human-induced actions.