Developing machine learning methods for seafloor habitat mapping utilizing high-resolution synthetic aperture sonar (SAS) data sets.
Remote sensing technologies have changed the way we are able to map and understand our planet. For ocean floor mapping, acoustic remote sensing technologies are the only effective methods for broadscale mapping of the seabed. Acoustic mapping technologies have advanced tremendously over the past two decades, and continue to improve, offering increasingly higher data resolution through more cost-effective field acquisition approaches. Relatively new acoustic remote sensing techniques, such as Synthetic Aperture Sonar (SAS), are now capable of mapping the seafloor environment down to 3cm horizontal resolution, with potentially game-changing applications in understanding the seafloor environment and ecosystem processes. However, with increasing resolution and wider adoption of these tools, data volumes continue to increase, posing big data challenges for processing. Data analysis can be labour intensive, forming bottlenecks in the data processing workflow. This project will develop new machine learning/deep learning/AI methods for processing SAS data sets. A variety of machine learning approaches will be tested and evaluated to generate seafloor thematic maps (e.g. surficial geology, predicted species distribution maps) through integration of the seafloor acoustic data with ground-validation information from subsea photographs and video.