Machine Learning for Automated Detection of Shipwreck Sites from Large Area Robotic Surveys

From May 23 - June 3, an expedition team will use autonomous underwater vehicles (AUVs) and remotely operated vehicles (ROVs) to explore Lake Huron’s Thunder Bay National Marine Sanctuary. They will generate large datasets for training and validating machine learning models for shipwreck detection using sonar imagery.

During the first year of their expedition project, the team of scientists will work on NOAA Great Lakes Environmental Research Laboratory’s Research Vessel Storm and deploy robotic systems to collect sonar data in regions with a high density of known shipwreck sites and exploratory survey regions which they have identified as having high potential for shipwreck discovery. The data collected will then be used to train and validate the team’s machine learning models with novel algorithms developed for shipwreck detection. With their ROV and AUV platforms equipped with cameras for recording images and videos of underwater sites, the expedition team will also assess the conditions of these targeted regions to guide future field expedition and exploration in Lake Huron.

Due to its maritime history and strategic location, Thunder Bay National Marine Sanctuary contains almost 100 known shipwreck sites and over 100 shipwrecks left to be found, thus offering a promising potential for both shipwreck discovery and development of machine learning algorithms. Shipwrecks help us better understand our past. But discovering and exploring them is expensive, time-consuming, and labor intensive. By advancing and training the capabilities of marine robotic systems to search for and survey shipwreck sites autonomously, the expedition team aims to increase the efficiency and decrease the costs associated with such exploration efforts.

Data collected and software developed throughout this project will not only inform and enhance public education and the management and conservation of important sanctuary resources, but also become widely applicable to the discovery of submerged maritime assets in the deep ocean. The data and software will be made publicly available and will serve as a benchmark for future re­search at the intersection of machine learning and ocean exploration.