June 29, 2017 — To monitor fish stocks in about 3 million square miles of ocean including the North Pacific Ocean and East Bering Sea, the Alaska Fisheries Science Center turned to facial recognition technology — or more accurately, “fishal recognition.”
Part of the National Oceanic and Atmospheric Administration’s National Marine Fisheries Service, the center is working to get more accurate counts of marine life by applying the same facial recognition techniques that identify people to recognize fish according to their facial features underwater.
The agency began experimenting with this technology several years ago and is now on its second generation of camera-based trawl, or CAM-Trawl. It has worked with ADL Embedded Solutions on hardware improvements, such as small form factor embedded Vision Boxes — the image acquisition computers — a Quad-Core Intel Core i7 processor to enable real-time processing of image data and waterproof enclosures and connectors.
Before using CAM-Trawl, the traditional method of measuring fish stocks had been to use trawlers to scoop up all the fish in a particular region of the ocean, bring them on deck, count them and then multiply that number out, according to said JC Ramirez, director of engineering at ADL. “If you did 10 square miles, multiply that by 10 for 100,” for example, he said.
But that method had several shortcomings, including extrapolation errors, high costs and ecological impacts on fish populations.
“They were looking for not only a more benign way to do that, but also perhaps a way that would allow them to cover more regions and get a little bit more specificity on the variations in the data from area to area,” Ramirez said.
Now researchers can pull a rig outfitted with the cameras that are linked back to ADL’s control computer, which typically resides in the wheelhouse. It monitors external sensors such as location, time, pressure sensors and radio frequency identification tag readings. It then, based on sensor input, communicates with one or more Vision Boxes to do the image capture, managing the remote on/off power, clock syncing, and starting and stopping of image collection.