AI will help in mass detection of prehistoric monuments in Poland

Scientists from Poland are the first in the world that plan to use an artificial neural network to detect pre-historic cemeteries, strongholds or settlements. "I expect a very large increase in data," says the project initiator, archaeologist Piotr Wroniecki from the University of Warsaw.

"We want to use artificial neural networks to detect ancient traces of man-made structures. In Poland there are still thousands of unknown relics of intense and long-lasting human presence," explained Piotr Wroniecki, whose team received a grant from the National Science Center for this purpose.

Artificial intelligence and learning neural networks are an innovative way to support human work from medical and technical diagnostics to automatic translations. For many years it was believed that machines should be fully programmed and all the knowledge should be written down in the form of code. However, the recently developed artificial neural networks work differently, more similar to a brain. As a result, systems based on them are capable of learning and improving their effectiveness “on their own”.

“Artificial neural networks need a very large amount of data to process, for example, satellite images, to properly work. For several decades, archaeologists have been analysing these images, which takes a long time for them. Additionally, people view images subjectively, often overlooking old traces of human activity. We hope that artificial intelligence will be able to mark them efficiently and objectively," Mr Wroniecki said.

The neural network will be programmed to focus on the so-called cropmarks visible on satellite or aerial photographs. The outlines of prehistoric houses are signalled, for example, by the presence of darker vegetation. New archaeological sites are often discovered by looking for characteristic changes visible from a bird's eye view. In archaeology, the artificial neural networks are already used to e.g. classify fragments of ceramics.

“Satellite photos will be uploaded to the artificial intelligence system. The system itself will find cropmarks, which it will classify as archaeological remains. Then it will outline them and mark them on the map. The archaeologist will then receive this "product".

"Manually this would take us weeks or even months. An artificial neural network will do it in several hours for hundreds of thousands of square kilometres," he argues.

"Until now, however, no one in the world has used artificial neural networks to analyse cropmarks in satellite imagery in search of traces of ancient human activity. In-depth analysis of satellite images will enable the analysis of traces of human presence in a comprehensive, holistic manner,” the archaeologist emphasized.

The project will last three years. The archaeologist hopes that it will create a prototype that can be used throughout Europe. For a decade, Wroniecki has been using non-invasive methods in archaeology on a large scale. Apart from aerial or satellite images, he also used various geophysical methods. Thanks to this, he discovered, among others, the lost city of Nieszawa ( Kuyavian-Pomeranian Province) and 7000 year old huge cult constructions.

“I hope that the system based on neural networks will bring us new revelations," Mr Wroniecki concluded.