Machine learning, neural networks

Application of neural networks based on transformer architecture in the archaeological data collections processing

A. V. Vokhmintcev^a, Kh. Mostafa^bc, V. R. Abbazov^c, M. A. Romanov^a, A. V. Mel'nikov<sup>cd

a</sup> Chelyabinsk State University, Chelyabinsk, Russia
^b Xuzhou Medical University, Xuzhou, China
^c Yugra State University, Khanty-Mansiysk, Russia
^d Ugra Research Institute of Information Technologies, Khanty-Mansiysk, Russia

Abstract: The paper presents algorithms for solving the task of 3d classification and instances segmentation of multimodal point clouds based on the multi-level graph convolutional network PointView-GCN and the transformer architecture Mask3d, respectively, which were used for detecting and structure decrypting an archaeological sites of the Bronze Age of the Southern Trans-Urals. Modifications have been made to an architecture of these models, which have improved the processing quality of sparse, uneven and noisy point clouds.

Keywords: earth remote sensing, methods methods of remote sensing of the Earth, instance segmentation, point clouds, transformers.

DOI: 10.14357/20718594250307