While oceans cover nearly three-quarters of Earth's surface, much of the underwater world remains difficult to see. Darkness and sediment limit the capabilities of traditional cameras, leaving researchers with an incomplete picture of what lies beneath the surface.
A new Office of Naval Research grant awarded to University of Miami electrical and computer engineering professor Shahriar Negahdaripour aims to improve how researchers use sonar imaging to explore underwater environments. Alongside doctoral student Guilherme Oliveira, the $700,000 award will support three years of research at the College of Engineering.
Unlike optical cameras, which rely on light, sonar systems create images using reflected sound waves. Because sound travels well through water and is less affected by murky conditions, sonar is one of the main tools used for marine research, target detection and localization, offshore infrastructure inspection, and search-and-rescue missions. These systems can reveal features that may be difficult to detect with conventional cameras under poor visibility.
Negahdaripour's research focuses on high-frequency forward-looking sonar, a technology capable of producing detailed two-dimensional images of underwater objects. His project will use machine learning to understand how sonar systems create images, helping researchers build more accurate three-dimensional models of underwater environments from two-dimensional sonar images.
"Sonar allows us to 'see' where optical systems fail, but there is still much to learn about how sonar images are formed and how we can extract more information from them," Negahdaripour said. "This project aims to help bridge that gap."
A major challenge in sonar research is that many commercial systems function as "black boxes." While the equipment produces detailed images of underwater environments, the proprietary hardware and processing software used to create those images are often not available to researchers. Sonar images can also vary from one system to another, making it difficult to fully understand how acoustic signals are transformed into images. By learning the process, researchers can better understand sonar data and improve the accuracy of sonar-based mapping systems.
For example, a sonar system may capture an image of a shipwreck on the seafloor. But researchers may not fully know how the sonar system's software translated the acoustic signals into the final picture. By learning the process researchers can better understand sonar data and improve the accuracy of sonar-based mapping systems. Negahdaripour's project hopes to overcome that challenge by developing a machine learning framework capable of reverse engineering three-dimensional object models from the images of a sonar system.
"Building accurate representations of underwater environments, from natural seafloor terrain to man-made offshore infrastructure, is transformational technology that could enable the creation of digital twins, virtual models that mirror real-world underwater environments," Negahdaripour said. "These tools could help researchers monitor marine ecosystems and inspect offshore infrastructure, turning the survey data into dynamic, living replicas of the ocean."
