This article was originally posted in August 2019. It has been re-posted here following a server issue in which the original post was accidentally removed.
Chennu A, Färber P, De’ath G, de Beer D, Fabricius KE (2017). A diver-operated hyperspectral imaging and topographic surveying system for automated mapping of benthic habitats. Sci. Rep. 7:7122. https://doi.org/10.1038/s41598-017-07337-y
Marine ecosystems such as coral reefs are currently being bombarded on all fronts by pollution, overfishing, and a rise in water temperature and acidity. 25% of reefs are in danger of extinction–and another 25% will be under threat by 2050. How can we surveil these diverse natural areas as they respond to human threats? Researchers from the Max Planck Institute in Germany have developed a new system–based on sophisticated imaging technology and artificial intelligence–which promises to revolutionize underwater mapping.
Counting (Underwater) Sheep
“Surveys…are a very time-consuming process because you cannot dive forever, afterwards you have to go look through all your pictures, through all your film material, through all your notes… And so it can take hours and hours of just trying to find out how healthy is your reef.” – Dr. Joost den Haan, HyperDiver team
Coral reefs teem with life and contain a higher density of species than any other marine environment. They provide important economic and environmental services such as fisheries, tourism, and coastal protection, that together are worth $30 billion annually. Monitoring these diverse areas is necessary for conservation and resource management–but their complexity makes this necessary monitoring difficult. Traditional coral reef surveys require expert ecologists to manually count and identify organisms while scuba diving. This laborious process covers a limited area, and provides only a small snapshot of reef life.
In recent years, reef-monitoring scientists have begun to capitalize on modern imaging technology. With image-based surveys, scientists can cover much larger areas and obtain a longer-lasting record of the reef. This also removes the necessity of on-the-fly interpretation, as scientists analyze the captured images post-dive. They can then use survey images to create detailed maps of what is living on the reef, and how healthy the reef is overall.
Hyperspectral Imaging (in a Nutshell)
Spectroscopy is a powerful method for identifying a material and defining its properties–all by examining how light interacts with it. It works because different materials have different light signatures, or “fingerprints”. Picture (pun intended) the digital camera in your phone–it shoots an image in three wavelength bands (corresponding to the red, green, and blue colors) which match the human vision. Each of these captured colors is like a page in a book, and the resulting selfie image we see is a combination of these three pages. In contrast, a hyperspectral camera records the target in hundreds of wavelengths across the electromagnetic spectrum. The resulting image contains not just three pages, but enough to rival a Harry Potter book. As a result, a hyperspectral image contains much more detailed information.
Enter the HyperDiver
The HyperDiver system is a novel diver-operated technology for surveying shallow marine ecosystems. At its heart is a hyperspectral camera for capturing high-resolution images of the seafloor. It also contains a suite of other sensors to measure other environmental parameters such as water acidity and dissolved oxygen levels. The system is battery-powered with up to 10 hours/use, and can be adjusted to be neutrally buoyant (weightless) in water.
To conduct a survey with the HyperDiver, a (human) diver swims along a pre-defined transect while pushing the instrument with the camera facing the bottom. A user interface aids the diver during the swim by displaying altitude and heading. During a pilot study of coral reefs in Papua New Guinea, the team was able to cover areas of 15-30 m2 per minute, collecting about 1.5 GB of hyperspectral data in a typical 3 minute transect.
HyperDiver missions generate a hefty amount of data–and since the whole point of this monitoring effort is to repeat surveys in as many places and as frequently as possible, analyzing all of these images would be impractical. So, why not get a machine to do it? This is precisely the idea behind the HyperDiver team’s machine learning technique approach. To transfer human knowledge to the computer system, human experts must first manually compare scanned images with a high-definition video to create a library of light signatures. Then, they use this database to “teach” the HyperDiver system to identify different life forms. Eventually, the system gets smart enough to analyze images on its own–producing automated maps and reports.
Mapping the Future
The ocean takes thousands of years to evolve naturally. In contrast, humans are inducing large, abrupt changes, with ripple effects that are already being seen in marine ecosystems. In order to react quickly to preserve marine species and manage resources, it is crucial for scientists to obtain accurate, real-time information. The HyperDiver system is an effective tool for monitoring and protecting important areas of the ocean. ■
I am a Ph.D. candidate at Boston University where I am developing an underwater instrument to study the coastal ocean. I have a multi-disciplinary background in physics and oceanography (and some engineering), and my academic interests lie in using novel sensors and deployment platforms to study the ocean. Outside of my scholarly life, I enjoy keeping active through boxing and running and cycling around Boston.