There are many applications for Autonomous Seaborne Vessels (ASVs). The seaborne cargo shipping industry moves over 9 billion tons of cargo per year, is worth $375 billion, and is responsible for 90 percent of world trade. Autonomous cargo ships could reduce the operating expenses of cargo ships by 44%. ASVs can also be used by the military for surveillance, and for autopilot of pleasure ships.
One of the main obstacles to the development of ASVs is that they need to obey the International Regulations for Preventing Collisions at Sea 1972 (COLREGs). COLREGs governs when a vessel has the right of way over other vessels, and the rules depend on the kind of ship encountered. For example, a motorized vessel must give way to a sailing vessel and vessels engaged in fishing. To follow these rules, it is necessary for an ASV to categorize other vessels. An ASV may need to classify ships for other reasons as well. For example, a military ASV may need to categorize hostile military vessels to determine how to best escape conflict.
Gnosptic Fields for Maritime Imagery are capable of accurately classifying ships into 70 fine-grained categories and seven categories that are relevant to COLREGs. The software takes images of ships as input and classifies the ship at multiple levels, e.g., a large ship that is also a cargo vessel. The software accomplishes this using Gnostic Fields, a brain-inspired classification algorithm for images.
The software is implemented in MATLAB. The algorithm can operate with a limited amount of data. When trained, it can classify ships quickly on a modern system equipped with a GPU (i.e., about 14 images per second in a MATLAB implementation).
This work was done by Christopher Kanan of Caltech for NASA’s Jet Propulsion Laboratory.