Autonomous underwater vehicles (AUVs) armed with mounted cameras and artificial intelligence (AI) for data processing can help map the species that live on the ocean floor. To create the last part of that vision, roboticists and marine scientists recently tested a new computer vision (CV) system to identify seafloor creatures.
The research was part of the Deep Links project, which is led by the University of Plymouth. Researchers from Oxford University, British Geological Survey, and the Joint Nature Conservation Committee also collaborated on the project. Deeps Links is funded by the Natural Environment Research Council (NERC).
“Autonomous vehicles are a vital tool for surveying large areas of the seabed deeper than 60 m (the depth most divers can reach). But we are currently not able to manually analyze more than a fraction of that data. This research shows AI is a promising tool but our AI classifier would still be wrong one out of five times, if it was used to identify animals in our images,” Ph.D. student Nils Piechaud, lead author on the study, says.
Piechaud adds, “This makes it an important step forward in dealing with the huge amounts of data being generated from the ocean floor, and shows it can help speed up analysis when used for detecting some species. But we are not at the point of considering it a suitable complete replacement for humans at this stage.”
UK’s Autosub6000 took to the sea in May 2016 in the North East Atlantic. During a single dive to 1,200 m below the ocean, the AUV gathered more than 150,000 images. A team manually analyzed 1,200 images from the bunch, “containing 40,000 individuals of 110 different kinds of animals (morphospecies), most of them only seen a handful of times,” according to the University of Plymouth. The accuracy of manual annotation ranged from 50 percent to 95 percent.
For the rest of the photos, the team used Google’s open access library called Tensorflow. There, a convolution neural network (CNN) was trained to recognize these deep-sea creatures, which neared humans’ high-end accuracy percentage at 80 percent. In addition, it also delivered a speed and consistency advantage.
The system can also reach up to 93 percent accuracy when identifying a specific species, that is, if the algorithm is fed a sufficient amount of data.
Instead of totally replacing manual annotation, the study looks to use a combination of AUVs and human specialists. AI tools can help speed up deep-ocean research and understanding, enhance data analysis, and assess prediction reliability.
The study was outlined in the article “Automated identification of benthic epifauna with computer vision,” published in Marine Ecology Progress Series.
