Tiny marine fossils called foraminifers or forams have been instrumental in guiding scientists studying global climate through the ages. The earliest record of their existence, which is evident from the millimeter-wide shells they leave behind when they die, dates back more than 500 million years. These single-celled protists thrived in many marine environments in their heyday; so much so that many sediments on the seafloor are made up of their remains.
Diverse and complex, the shells can provide valuable information about the state of the ocean, along with its chemistry and temperature, during the lifetime of the forams. But until now, the process of identifying, cataloging and classifying these microscopic organisms has been a tedious task for research labs around the world.
Now there is hope that simple jobs can be delegated to a more mechanical workforce in the future. A team of engineers from North Carolina State University and the University of Colorado Boulder has built a robot specifically designed to isolate, view, and classify individual forams by species. It’s called Forabot, and it’s made from off-the-shelf robotic components and proprietary artificial intelligence software (now open source). In a small proof-of-concept study published this week in the journal Geochemistry, Geophysics, GeosystemsThe technology’s identity accuracy was 79 percent.
“Due to the small size and great abundance of planctic foraminifers, hundreds or possibly thousands of them can often be collected from a single cubic centimeter of ocean floor mud,” the authors wrote in their paper. “Researchers use the relative abundances of foram species in a sample, as well as determining the stable isotope and trace element compositions of their fossilized remains, to gain insight into their paleoenvironment.”
[Related: Your gaming skills could help teach an AI to identify jellyfish and whales]
However, before any formal analysis can be done, foraminifera must be extracted. This is where Forabot can come in. After the scientists wash and sift the samples filled with sand-like shells, they place the materials in a container called an isolation tower. From there, the single forams are transferred to another container, called the imaging tower, where an automated camera captures a series of images of the sample and is then fed to AI software for identification. After the sample is sorted by the computer, it is taken to a sorting station, where it is distributed into a suitable well according to the species. In its current form, Forabot can distinguish six different types of forams and can process 27 forams per hour (the researchers’ quick calculation shows that it can examine about 600 fossils per day).
For the classification software, the team modified a neural network called VGG-16 that was pre-trained on more than 34,000 images of planktonic forams collected worldwide as part of the Endless Forams project. “This is a proof-of-concept prototype, so we’re going to increase the number of foram species it can identify,” Edgar Lobaton, an associate professor at NC State University and co-author of the paper, said in a press release. release. “We are also optimistic that we can improve the number of forams it can process per hour.”
Watch Forabot in action: