Remote Research of Archaeological Sites of the Southern Trans-Urals Using Geophysics and Machine Learning

The so-called “Country of Cities” discovered in the second half of the 20th century in the Southern Trans-Urals — more than two dozen fortified settlements of the Middle Bronze Age belonging to the Sintashta culture (about 3–2 thousand years BC) is a unique object of interdisciplinary research. In this paper the study of the architecture of these settlements is carried out by interpreting aerial photographs, space photographs and geophysical methods: magnetometry and areal electromagnetic profiling with the AEMP-14 induction system. The construction of orthophotoplans and a digital relief model is made based on UAV survey data and ground surveys using GNSS and tacheometry. Fundamentally new opportunities are provided by applying modern methods of detection, classification and segmentation of objects based on the use of convolutional neural networks to the obtained data. This paper presents and discusses the results of applying neural networks based on graphs and transformer architecture to the problem of 3d archaeological sites segmentation and methods of their detection based on residual neural networks and networks with transformer architecture.