Application of shallow geophysical methods and machine learning for detecting remains of early medieval settlements in south-eastern Poland.

Poland's landscape is a testament to its deep-rooted agricultural history, characterized by ancient field systems echoing the spatial layouts in Celtic fields throughout Europe. These intricate and expansive layouts pose a significant challenge for archaeologists and researchers dedicated to uncovering the secrets of the past. The focus of this study is to meticulously explore and analyze these extensive field systems, which often cover large areas and require a detailed and systematic approach. To navigate this complex task, researchers employed cutting-edge deep learning neural networks (DLNN), particularly the U-Net model. This approach involved semantic segmentation of data derived from Airborne Laser Scanning (ALS) to automate the identification of these significant archaeological sites. The team successfully identified hundreds of ancient sites across Poland by harnessing the power of ALS data combined with thorough desk-based analysis.

The research concentrated on specific sites in southern Poland, namely in the areas around Trzebinia and Jaworzno. Various geophysical methods were utilised here, including Magnetometry and the Slingram Electromagnetic Induction Method. These techniques aimed to confirm the existence of preliminary archaeological features in the region. The researchers conducted Magnetic Gradiometry and Electromagnetic Measurements across different terrains, including cultivated fields and forests. They specifically targeted relict embankments that once delineated old fields. The findings from these investigations were striking. The geophysical profiles of the two studied areas revealed significant differences. In the first area in a current crop field, researchers observed point-like, strong anomalies in both vertical magnetic gradient and electrical conductivity. In contrast, the wooded study area exhibited weaker but continuous anomalies, suggesting the presence of buried burnt clay formations. A key aspect of this research was integrating Ground Conductivity assessments with vertical magnetic gradient evaluations. This approach was crucial in correlating data from both methods. At the first site, variations in conductivity at different depths hinted at geological transitions or man-made structures beneath the surface. Meanwhile, at the second site, resistivity patterns suggested an anthropogenic alteration of water conditions, possibly resembling an artificial fault.

Integrating a machine learning system into this research process marked a significant advancement. It facilitated the automated segmentation of ALS data, greatly enhancing the efficiency of detecting and mapping cultural resources over large areas. Combined with traditional geophysical methodologies, this innovative approach provided a non-invasive means of identifying potential archaeological objects. This was crucial for the effective management and preservation of heritage sites. In summary, this comprehensive interdisciplinary study represents a fusion of advanced technological solutions with traditional geophysical methods. It offers valuable new insights into detecting and interpreting archaeological features, potentially revolutionizing the field of archaeological exploration and heritage conservation. The research highlights the importance of integrating diverse methodologies to uncover the intricacies of our past, ultimately contributing to a deeper understanding of human history and its impact on the landscape.