Development of a platform for automated detection of marine fauna using machine learning

• Grant number: 25/06265-3
• Support Opportunities: Scholarships in Brazil - Master
• Start date: August 01, 2025
• End date: March 31, 2027
• 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
• Associated research grant: 19/12551-8 - Benthic connections of high Southern Latitudes: BECOOL, AP.PFPMCG.TEM

Abstract

Monitoring marine biodiversity is essential for ecosystem conservation, but traditional methods of biological sampling can be invasive and harmful to species. Technological advances, such as the use of remotely operated vehicles (ROVs), have enabled the adoption of non-invasive approaches to studying marine ecosystems. However, these operations generate large volumes of data, which require significant effort for the manual identification and annotation of recorded organisms. In this context, Artificial Intelligence (AI) stands out as an effective solution for the automated analysis of videos, offering a fast and accurate alternative. The application of this technology is particularly important in ecologically significant ecosystems, such as deep-sea coral banks and Antarctic marine environments, both of which are vital to ocean health and highly vulnerable to anthropogenic impacts. The main objective of this project is to develop an interactive graphical interface for video annotation, integrated with an AI model for the automated detection and classification of marine organisms. After its development, the platform will be tested using videos from the Southern Ocean and deep-sea coral banks in the Campos Basin. The data to be analyzed were obtained from videos captured by ROVs during the DECODE project in the Campos Basin and the BECOOL project in Antarctica, allowing the tool's effectiveness to be tested in different ecological contexts. Differences in species composition and abundance-both among coral banks and between AI-generated data and traditional annotations-will be statistically tested using Non-metric Multidimensional Scaling (nMDS) and Permutational Multivariate Analysis of Variance (PERMANOVA). The integration of AI into the monitoring of these ecosystems presents an innovative and promising approach that, in addition to contributing to the advancement of knowledge on local biodiversity, will also significantly reduce the time required for video analysis in this field of research. (AU)