20.500.12556/RUL-173522 Uporaba globokega učenja za analizo zelo visoko ločljivih posnetkov v ArcGis Pro na območju Slovenije magistrsko delo Using deep learning for analyzing very high-resolution images in ArcGIS Pro for the area of Slovenia master thesis Zaradi napredka na področju globokega učenja za obdelavo posnetkov in prepoznavanja vzorcev je klasifikacija podatkov daljinskega zaznavanja v zadnjih letih močno napredovala. Prednaučeni modeli globokega učenja obljubljajo hitro in avtomatizirano analizo prostorskih podatkov, vendar njihova uporabnost v novih prostorskih in vsebinskih kontekstih pogosto ostaja neraziskana. V nalogi predstavljamo oceno učinkovitosti uporabe štirih prednaučenih modelov, integriranih v okolje ArcGIS Pro, na digitalnih ortofotih Slovenije iz let 2018 in 2021. Testirani so bili modeli Building Footprint Extraction – USA, Segment Anything Model (SAM), Tree Segmentation in Road Extraction – North America. Analiza je bila izvedena na dveh prostorsko in vsebinsko različnih območjih, urbanem (naselje Brdo v Ljubljani) in ruralnem (občina Ig). Rezultati kažejo, da prilagajanje ključnih parametrov lahko izboljša učinkovitost modelov, vendar pri uporabi modelov na ortofotih ostajajo omejitve, povezane s specifičnostjo lokalnega prostora. Najvišja točnost je bila dosežena pri prepoznavanju stavb, medtem ko so modeli za segmentacijo dreves in cest pogosteje generirali napačne zaznave. Rezultati kažejo, da so prednaučeni modeli primerni kot izhodišče za hitro obdelavo posnetkov zelo visoke ločljivosti, vendar je za doseganje višje natančnosti potrebno dodatno učenje in ponovni trening na lokalnih podatkih. Thanks to advances in deep learning for image processing and pattern recognition, remote sensing data classification has progressed significantly in the last few years. Pre-trained deep learning models promise fast and automated analysis of spatial data, but their applicability in new spatial and content contexts often remains unexplored. In this study, we evaluate the performance of four pre-trained deep learning models integrated into ArcGIS Pro on Slovenia’s national digital orthophotos (DOF) from 2018 and 2021. The models tested were Building Footprint Extraction Model – USA, Segment Anything Model (SAM), Tree Segmentation Model and Road Extraction – North America Model. The analysis was conducted for two spatially and contextually different areas, one urban (the settlement Brdo in Ljubljana) and one rural (the municipality of Ig). Results demonstrate that tuning key parameters can improve model performance, but that inherent limitations persist when applying these models to local orthophotos. The highest accuracy was achieved for building footprint extraction, whereas the tree and road segmentation models exhibited higher rates of false detections. Overall, pre-trained models offer a valuable starting point for processing very high resolution (VHR) imagery, but attaining higher precision requires additional fine-tuning and re-training on locally representative data. magistrska dela globoko učenje prepoznavanje objektov Slovenija ArcGIS Pro prednaučeni modeli master thesis deep learning object detection Slovenia pre-trained models Building Footprint Extraction Segment Anything Tree Segmentation Road Extraction true false false [M. Gelebeshova] Slovenski jezik Angleški jezik Magistrsko delo/naloga 2025-09-18 08:45:32 2025-09-18 08:45:35 2025-11-10 03:41:03 0000-00-00 00:00:00 2025 0 Ljubljana 2025 Ljubljana 1 spletni vir (1 datoteka PDF (IX, 37 str., [14] str. pril.)) 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 528:004.81(497.4)(043.2) 174284 249545987 19545.pdf 19545.pdf 1 793B1F4B8F3F9AE6F2E722AF8BDDFFFF 2316eb7f9f6e9281373a3860bfd83a1b81c89eea1bc3aa6227545120cf4a6e72 e7a7df03-945a-11f0-9328-0050569b8976 https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=218038 Fakulteta za gradbeništvo in geodezijo 0 0 0