Scene Understanding for Underwater Imagery

Overview

Underwater ecosystems, including coral reefs and seagrass meadows, play a critical role in maintaining marine biodiversity, providing coastal protection, and supporting fisheries and tourism economies that millions depend upon globally. These habitats are increasingly vulnerable to climate change, pollution, and other anthropogenic impacts, demanding urgent efforts to monitor and restore them. Accurate scene understanding of underwater imagery enables fine-scale ecosystem monitoring across spatial and temporal scales, supporting essential activities such as habitat and biodiversity assessment, validation of aerial and remotely sensed data, and targeted reef restoration interventions.

We are looking for passionate and diligent PhD students to work with us on challenging, real-world tasks in underwater perception. Whether you are driven by marine conservation and making a difference, or if you're passionate about designing cutting-edge AI for complex environments, we'd love to hear from you!

This project will build upon our prior work on underwater perception: Weakly Supervised Segmentation of Underwater Imagery.

Underwater perception

Underwater perception sits at the intersection of artificial intelligence, computer vision, and marine science, focusing on developing algorithms and models to analyse complex underwater imagery collected by robotic vehicles or divers. Underwater perception tackles the unique challenges of working with underwater images, which are often constrained by limited and difficult-to-label datasets and distorted by characteristics such as turbidity and low light. The task involves accurately detecting, classifying, and mapping marine habitats, conditions or species.

What we offer

The QUT Centre for Robotics is Australia’s top-ranked research hub for robotics and is dedicated to transforming innovative ideas into practical applications with real-world impact. The centre embraces diversity in an exciting, collaborative environment with a vibrant culture. As all PhD students are supervised by at least two experienced academics, you would collaborate with leading researchers and craft your mark in the realms of AI and robotics. The Centre for Robotics is based at the QUT Gardens Point campus, beautifully located adjacent to Brisbane's Botanic Gardens and the CBD.

Our past collaborations have included the CSIRO’s Data61 and Oceans and Atmosphere divisions, and we are currently collaborating with the Reef Restoration and Adaptation Program and the Australian Institute of Marine Science.

Research activities

Research Program 1: Scene Understanding for Continuous Variables in Underwater Imagery

The first research program focuses on advancing scene understanding of continuous variables in underwater imagery to capture subtle ecological indicators, such as gradients between life and death in coral recruits or varying stages of coral bleaching. Unlike traditional classification approaches, which often apply rigid labels (e.g., “alive” or “dead”), this research aims to develop models that can detect and interpret intermediate states within these ecosystems. Such nuanced scene understanding is crucial for tracking ecosystem health, particularly as early signs of stress or partial recovery in corals can be indicators of broader environmental changes. By creating methods capable of recognising these gradients, we can improve our ability to monitor reef resilience, support timely interventions, and contribute valuable data for conservation and restoration efforts.

Research Program 2: Coral Device Re-Identification with Geographic Priors

The second research program focuses on coral device re-identification, specifically tracking coral ceramic devices used in reef restoration to re-seed areas with temperature-resilient coral species. Over time, these devices can become obscured as algae and coral naturally grow over their surfaces, challenging traditional identification methods. This research aims to develop algorithms capable of recognising partially obscured devices by leveraging both visual features and the approximate GPS locations recorded at deployment. By combining image-based re-identification with spatial data, we can enhance monitoring accuracy, track growth and survival of transplanted corals, and support data-driven reef restoration efforts.

Research Program 3: Open-Set Recognition of Fine-Grained Underwater Species

The ocean can be considered as an open-set problem due to its vast biodiversity, encompassing countless species and ecological features, many of which remain undocumented or only partially understood. This project explores open set recognition in underwater species identification, aiming to detect and classify marine organisms that the model may not have encountered before. Unlike traditional models limited to predefined classes, open set recognition leverages techniques that can recognise novel species or unusual features in marine imagery. This research program could involve the use of large language models and ecologist-provided natural language descriptions to enable detection of previously unseen species.

Outcomes

This project aims to contribute novel approaches for scene understanding of underwater imagery, which could have outcomes for monitoring marine ecosystems, reef restoration and adaptation, and quantifying impacts on biodiversity.

The technical contributions could also have implications for other challenging tasks in land-based environmental monitoring or agriculture.

Skills and experience

To be considered for this project, you’ll need a strong academic background. Proficiency in programming and a passion for artificial intelligence, computer vision and/or environmental conservation is desirable but not essential.

Qualifications

• A bachelors or master degree in robotics, computer vision, computer science, mathematics, mechatronics, marine science, environmental science or related areas.
• A strong background, or an interest in, machine learning, deep learning, computer vision, or robotics.
• Proficiency, or an interest in, programming languages such as Python, C++, MATLAB or similar.
• Strong analytical and problem-solving skills.
• Strong written and verbal communication skills.

Diversity

We highly encourage applications from underrepresented groups, including Women in STEM and First Nations peoples. The QUT Centre for Robotics makes safety, inclusivity and support a priority so that staff and students have the best possible chance to succeed. Many Centre members, including both members of the supervisory team, are part of the QUT Ally Network and are trained to understand sexuality and gender issues.

Scholarships

You may be eligible to apply for a research scholarship.

Explore our research scholarships

Keywords

• Computer Vision
• Marine Science
• Robotics
• Artificial Intelligence
• Underwater Perception
• Environment

Contact

Contact the supervisor for more information.