Bedrock Ocean Exploration now runs automated target recognition on board its fleet of autonomous underwater vehicles, the company announced. The model finds and classifies objects on the seafloor, such as debris, infrastructure and possible unexploded ordnance, as each survey line is finished.

When a vehicle finishes a line, the model usually runs in under a minute. When the vehicle surfaces, it sends the contacts and their positions ashore, each with its sidescan sonar image, and experts accept, reject or annotate them while the vehicles are still on site. This usually takes under five minutes, against days to months when analysts pick contacts by hand after a survey.

In mine countermeasures, for example, someone is deciding whether an object could be explosive, so a person needs to be in the loop in real time, said Bridgit Boulahanis, head of data and solutions at Bedrock. The model is trained on Bedrock's own sidescan sonar data, supplemented by open-source datasets, and the experts' decisions are used to keep retraining it.

Source: Seapower

Background

Bedrock Ocean Exploration

Ocean survey company

Ocean survey company based in Richmond, California, with a fleet of autonomous underwater vehicles.