Researchers from the Virginia Institute of Marine Science at the College of William and Mary, Williamsburg, Va., have developed an artificial neural network (ANN) that works with an autonomous underwater vehicle named Fetch. The idea is to eventually train the vessel's computer to recognize underwater threats including submarines, missiles, mines, and even people. Currently, the AUV can only distinguish between different fish species. But scientists think it's a good start toward patrolling coastlines, harbors, and moored naval vessels.
Researchers trained the ANNs by grouping side-scan sonar data of distinctive fish species into test sets. They also used enhancement algorithms and image processing to teach the computer to identify qualities of different fish species. According to a report in New Scientist, Fetch recognized two marine fish species - jacks and sharks. "It's amazing how well this particular type of neural network works with noisy data," says project leader Mark Patterson, associate professor of marine science. "In the future, we hope to expand the classifier's library to include dozens of species."