Instruments and Systems: Monitoring, Control, and Diagnostics Annotation << Back Semantic Segmentation Algorithms for Multi-Temporal Multispectral Images Using Deep Learning YU.YU. GROMOV, I.N. ISHCHUK, YU.G. VESELOV, V.V. RODIONOV The article considers the problem of semantic segmentation of various objects, including vegetation, road surfaces, vehicles and engineering structures, with examples of their condition assessment. Vegetation biomass volume assessment is of great importance for effective biomass management, while qualitative assessment of vegetation biomass requires quantitative analysis of tree canopy. Automated extraction of tree canopy facilitates faster and more efficient assessment of biomass content in vegetative zones compared to traditional manual mapping methods. Road surface analysis allows distinguishing between asphalt and concrete surfaces, as well as calculating their areas. In turn, assessment of the materials from which vehicles are made makes it possible to distinguish automobile and armored transport from their models made of materials other than metal, as well as to identify objects made of aluminum, such as airplanes and helicopters. Experimental studies were conducted in an engineering town located in the Sovetsky district of the Voronezh urban district, with an area of 11.4 hectares. Using a multispectral camera of an unmanned aerial vehicle, multi-temporal images with a spatial resolution of 15 cm were obtained in four ranges: Blue, Green, Red and FIR. The earth's surface was surveyed 6 times during the day from 6.00 to 24.00 with an interval of 4 hours. The study attempted to automatically determine the parameters of objects using a deep learning convolutional neural network U-Net. The obtained object data were refined using spectral indices and by checking the segmentation results with field studies. The accuracy of object parameter estimates is consistent with field measurements and has shown high efficiency in segmenting objects in both small and large areas at a relatively lower cost. Keywords: multi-temporal multispectral imaging, unmanned aerial vehicle, deep learning, object parameter estimation, semantic segmentation. DOI: 10.25791/pribor.2.2025.1555 Pp. 01-16.