20.500.12556/RUL-173250 Samodejna segmentacija krošenj dreves v LiDAR podatkih: primerjava klasičnih in nevronskih metod Automatic tree canopy segmentation in LiDAR data: a comparison of classical and neural methods Ocena gozdnega stanja nam daje pomembne podatke o zdravstvenem stanju in biotski raznovrstnosti gozda. Ob enem je dolgotrajen proces, ki ga je treba redno izvajati. V diplomskem delu smo raziskali metode segmentacije posameznih dreves iz LiDAR podatkov, ki bi proces poenostavile in pohitrile. Testirali smo dve že razviti metodi segmentacije. Prva je temeljila na tvorjenju VMK (angl. Canopy height model) in rasti regij (angl. region growing). Druga je uporabljala model globokega učenja in se naslanja na arhitekturo U-Net. Metodi smo testirali na naših podatkih. Metodi sta dosegli 77\ \% natančnost. Pokazali smo, da je metoda globokega učenja uspešnejša za segmentacijo nižjih dreves in gostih iglastih gozdov. Utemeljili smo potencial globokega učenja za obravnavano nalogo. Assessment of forest condition provides important information about the health status and biodiversity of the forest. At the same time, it is a long-lasting process that needs to be carried out regularly. In this thesis, we investigated methods for segmenting individual trees from LiDAR data, which could simplify and accelerate the process. We tested two already developed segmentation methods. The first was based on generating a Canopy Height Model (CHM) and region growing. The second used a deep learning model and was based on the U-Net architecture. Both methods were tested on our dataset. The methods achieved an accuracy of 77\ \%. We showed that the deep learning method is more successful for the segmentation of smaller trees and dense coniferous forests. We substantiated the potential of deep learning for the considered task. segmentacija dreves lasersko skeniranje oblak točk konvolucijske mreže tree segmentation laser scanning point cloud convolutional neural networks true false false Slovenski jezik Angleški jezik Diplomsko delo/naloga 2025-09-15 10:25:05 2025-09-15 10:25:10 2025-11-07 03:46:18 0000-00-00 00:00:00 2025 0 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 38193 252879363 Bertoncelj_Miha_maksimiljan_-_Samodejna_segmentacija_krosenj_dreves_v_LiDAR_podatkih_primerjava.pdf Bertoncelj_Miha_maksimiljan_-_Samodejna_segmentacija_krosenj_dreves_v_LiDAR_podatkih_primerjava.pdf 1 EEE10BD794519A049D6EF3F6B5BDBB35 abbefa75e328e08a22d1287dd16c55f1966a79a645eece8df67419aa947f9c12 534ee265-920d-11f0-9328-0050569b8976 https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=217457 Fakulteta za računalništvo in informatiko 0 0 0