Georgian Scientists/ . 6 N 3, 2024 97 Georgian Scientists Vol. 6 Issue 4, 2024 https://doi.org/10.52340/gs.2024.06.04.10 , , tkeshelashvili.nika@gtu.ge, https://orcid.org/0009-0008-9353-6897 . , . , , dji Mavic 3 , 128 . , . Mask R-CNN . 4 , . , . , dji Mavic 3 3 , Mask R-CNN 92%- . Georgian Scientists/ . 6 N 3, 2024 98 , , , , , . , . : , , , , . 3D , , . (Audebert et al., 2018). , , , . . , , . , . , , . ( ) , . , ( ) Georgian Scientists/ . 6 N 3, 2024 99 (Colomina & Molina, 2014). , . (Calantropio et al., 2021). , , , . , . . , , . , , . 16.8 2 , 2014 1320 . , , , . 2024 , dji Mavic 3 . dji Mavic 3 (C1 ) , RGB . 1- . GNSS (GPS, Galileo, BeiDou). , , , , Emlid Reach RS 2+ GNSS Georgian Scientists/ . 6 N 3, 2024 100 . , . , 128 . 120 , - 8.5 . 1- . 1. Zone-2 1600 , , Agisoft Metashape Professional 1.8 . 4/3 CMOS, : 20 MP : 84° : 24 : f/2.8 - f/11 8-1/8000 . 5280×3956 1. dji Mavic 3 ( - 3 ) ( 2). Georgian Scientists/ . 6 N 3, 2024 101 2. Mask R-CNN (Mask Region-based Convolutional Neural Network) (He et al., 2017) . , Faster R-CNN , , . Mask R-CNN- mXm . : , ^ = ( + ) (1) , , ^ . , (2): Georgian Scientists/ . 6 N 3, 2024 102 = , log , ^ + , log (1 , ^ ), (2) , m , , - . . 2- . 2. Zone-1 ( 3), 330 . , Zone-1 Zone-3, , 70% / 30% , Zone-2 . , , , , , , . , 3 (RGB) , . 4 (RGBD) , , , . Mask R-CNN ResNet-101 0.001 4 70 128 x 128 0.9 0.0001 , , Georgian Scientists/ . 6 N 3, 2024 103 3. (Zone-1) 4 2- 128X128X1 . , , . Mask R-CNN , Tensorflow Keras V2.16.1 . 4- . Georgian Scientists/ . 6 N 3, 2024 104 4. : 1. RGBD ; 2. ; 3. 3- (Zone-2) . 92%- , . 3. , Zone-2 , „ “. 450- , , . 6- Zone-2 , . Precision 0.92 92% Recall 0.84 84% IoU 0.77 F1-Score 0.89 Mask R-CNN Georgian Scientists/ . 6 N 3, 2024 105 5. ( 4 5) , , , , , , . . , ( , ). , . , , . „ “ , . , , , , . . Georgian Scientists/ . 6 N 3, 2024 106 6. 6- Zone-2 . 16, . , 16 , ( 3 11), (4, 15) , . Mask R-CNN , 1 , 0 - ( 6). GIS- . , . . (Gribov, 2019) . Georgian Scientists/ . 6 N 3, 2024 107 7. 7- . , , , . , , , . dji Mavic 3 3 128 . Mask R-CNN , . 92%- . , , Georgian Scientists/ . 6 N 3, 2024 108 . , . 2. Audebert, N., Le Saux, B., & Lefevre, S. (2018). Beyond RGB: Very high resolution urban remote sensing with multimodal deep networks. ISPRS Journal of Photogrammetry and Remote Sensing, 140, 20-32. https://doi.org/https://doi.org/10.1016/j.isprsjprs.2017.11.011 Calantropio, A., Chiabrando, F., Codastefano, M., & Bourke, E. (2021). DEEP LEARNING FOR AUTOMATIC BUILDING DAMAGE ASSESSMENT: APPLICATION IN POST-DISASTER SCENARIOS USING UAV DATA. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, v-1-2021, 113-120. https://doi.org/10.5194/isprs-annals-V-1- 2021-113-2021 Colomina, I., & Molina, P. (2014). Unmanned Aerial Systems for Photogrammetry and Remote Sensing: A Review. ISPRS Journal of Photogrammetry and Remote Sensing, 92, 79-97. https://doi.org/https://doi.org/10.1016/j.isprsjprs.2014.02.013 Gribov, A. (2019). Optimal Compression of a Polyline While Aligning to Preferred Directions. International Conference on Document Analysis and Recognition Workshops (ICDARW), (pp. 98-102). Sydney, NSW, Australia. https://doi.org/10.1109/ICDARW.2019.00022 He, K., Gkioxari, G., Dollar, P., & Girshck, R. (2017). Mask R-CNN. IEEE International Conference on Computer Vision (ICCV), (pp. 2980-2988). Venice, Italy. https://doi.org/10.1109/ICCV.2017.322 Georgian Scientists/ . 6 N 3, 2024 109 UAVs and AI for Very High Resolution Map Creation and Automatic Object Detection Abstract Automation of object detection and extraction from VHR aerial imagery for high definition mapping is significant challenge for various remote sensing and photogrammetry applications. This study investigates the use of Unmanned Aerial Vehicles (UAVs) and Deep Learning models to enhance the traditional, manual methods and increase accuracy and efficiency of spatial data acquisition and processing. The dji Mavic 3 UAV was used to capture high-resolution aerial images over a 128 hectare area in town Tsageri and village Chalistavi, located in the Racha-Lechkhumi and Lower Svaneti region, Georgia. One of the main objective of this research was to detect and extract building footprints from generated orthomosaics using deep learning techniques. The Mask Region-based Convolutional Neural Network (Mask R-CNN) was employed for this task. To further improve the model's performance, Digital Elevation Models were integrated with the orthomosaics, resulting in a four-channel raster that provided additional topographical context for more accurate building extraction. The findings indicate that the integration of Unmanned Aerial Vehicles and Deep Learning technics offers significant improvements in acquisition and processing efficiency of geospatial data. Namely, we were able to create VHR orthomosaics with spatial resolution of 3.2 cm using dji Mavic 3 UAV. Trained Mask R-CNN model showed 92% precision for building prediction in the study area. Methods and technics described in this paper have important implications for various applications, including land administration, infrastructure development and environmental management. Methodology used in present study provides a valuable framework for future researches and practical applications, especially in such developing and mountainous countries as Georgia, where high-resolution and accurate spatial data is often unavailable due to complex environmental conditions and economic factors. Keywords: UAVs, DEM, Orthomosaic, Artificial Intelligence, Automatic object detection