Machine Learning Approaches to Classification of Seafloor Features from High Resolution Sonar Data

Wednesday, 17 December 2014
Denson glen Smith1, Lawson Ed2, Don Sofge2, Paul A Elmore3 and Fredrick Petry3, (1)University of New Orleans, New Orleans, LA, United States, (2)Naval Research Lab DC, Information Technology Division, Washington, DC, United States, (3)Naval Research Lab Stennis Space Center, Marine Geosciences Division, Stennis Space Center, MS, United States
Navigation charts provide topographic maps of the seafloor created from swaths of sonar data. Converting sonar data to a topographic map is a manual, labor-intensive process that can be greatly assisted by contextual information obtained from automated classification of geomorphological structures. Finding structures such as seamounts can be challenging, as there are no established rules that can be used for decision-making. Often times, it is a determination that is made by human expertise. A variety of feature metrics may be useful for this task and we use a large number of metrics relevant to the task of finding seamounts. We demonstrate this ability in locating seamounts by two related machine learning techniques. As well as achieving good accuracy in classification, the human-understandable set of metrics that are most important for the results are discussed.