The Brief study is brief information about interesting academic work.
The big idea
In collaboration with the US Navy Branch for underwater archeologyI learned a computer how to recognize shipwrecks on the ocean floor from scans made by planes and ships on the surface. The computer model that we created is 92% accurate in detecting known shipwrecks. The project focuses on the coasts of the continental United States and Puerto Rico. It is now ready for use to detect unknown or unreported shipwrecks.
The first step in creating the shipwreck model was to teach the computer what a shipwreck looked like. It was also important to teach the computer how to distinguish the debris and topography of the seabed. To do this, I needed many examples of shipwrecks. I also had to learn the model of what the natural ocean floor looked like.
It is convenient for the National Oceanic and Atmospheric Administration to maintain public database of shipwrecks. It also has a large public database with various types of images collected from around the world, including sonar and leader images of the seabed. The images I used extend just over 14 miles (23 kilometers) from shore and to a depth of 279 feet (85 meters). This image contains huge areas without shipwrecks, as well as accidental shipwrecks.
Why it matters
Finding shipwrecks is important for understanding the human past – think of trade, migration, war – but underwater archeology is expensive and dangerous. A model that automatically maps all shipwrecks over a large area can reduce the time and expense required to search for wreckage, whether with underwater drones or divers.
The Navy’s Underwater Archeology Club is interested in this work, as it can help the unit find undiscovered or unknown shipwrecks. More broadly, this is a new method in underwater archeology that can be extended to look for different types of submerged archaeological features, including buildings, statues and airplanes.
What other research is being done in this area
This project is the first model focused on archeology, which is built to automatically identify shipwrecks over a large area, in this case the entire coast of the continental United States. There are several related projects that focus on finding shipwrecks using deep study and images collected by an underwater drone. These projects can find a handful of shipwrecks that are in the area immediately around the drone.
We would like to include in the model more data on shipwrecks and images from around the world. This will help the model to recognize many different types of shipwrecks. We also hope that the Navy’s underwater archeological branch will sink to some of the places where the model found shipwrecks. This will allow us to check the accuracy of the model more carefully.
I am also working on several other archaeological machine learning projects and they are all based on each other. The overall goal of my work is to create a personalized model for archaeological machine learning. The model will be able to quickly and easily switch between predicting different types of archaeological features, both on land and underwater, in different parts of the world. To this end, I am also working on projects aimed at finding ancient Mayan archaeological structures, caves of Mayan archeological sites and Romanian tombs.
Leila Character, PhD student in geography, University of Texas at Austin College of Liberal Arts