Instruments and Systems: Monitoring, Control, and Diagnostics Annotation << Back Using Segmentation Neural Networks for Creating Models of the Underlying Surface K.V. PARFENTIEV, K.A. NEUSYPIN The paper considers the problem of constructing underlying surface models using the sensors of various spectral ranges. For solving this problem a new approach based on the using of boundary detection methods is proposed. For correction of the obtained models neural network structure is proposed. It consists of 58 layers and including convolutional, pooling, fully connected and segmentation layers. The temporal characteristics of the neural network model are given, on the basis of which it is concluded that the results can be used for solving the problem of determining navigation parameters. Keywords: neural networks, unmanned aerial, photogrammetry, underlying surface, edge detection. DOI: 10.25791/pribor.7.2022.1347 Pp. 13-17.