Detekcija klizišta na LiDAR snimkama temeljena na dubokom učenju

master's thesis

• Title (english): Deep Learning-based Landslide Detection in LiDAR Images
• Author: Luka Šimac
• Mentor: Marko Subašić (mentor)
• Committee member: Marko Subašić (predsjednik povjerenstva)
• Committee member: Juraj Petrović (član povjerenstva)
• Committee member: Krešimir Križanović (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2022-07-12, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Tema diplomskog rada je detekcija klizišta na LiDAR snimkama temeljem dubokog učenja. Za rješavanje problema korišteni su algoritmi temeljeni na semantičkoj segmentaciji i dubokim konvolucijskim mrežama. Za kvalitetno rješavanje problema potrebno je bilo provesti predprocesiranje izvornih LiDAR snimki kako bi se mogao izvući maksimalan broj informacija i kreirati adekvatan skupa podatka za treniranje i evuliranje modela te isto tako provesti agumentaciju tokom treniranja modela u svrhu povećanja količine podataka. Istraživanjem dostupnih algoritama semantičke segementacije odlučeno je da su mreže arhitekture enkoder-dekoder pogodne za problem kojim se bavimo. U svrhu dobivanja kvalitetnih rezultata proveden je niz eksperimenata gdje su se ispitivale različite konfiguracije enkoder-dekoder arhitekture u jednakim uvjetima.

Abstract (english)

The topic of the thesis is the detection of landslides on LiDAR images based on deep learning. Algorithms based on semantic segmentation and deep convolutional networks were used to solve the problem. To solve the problem well, it was necessary to pre-process the original LiDAR images in order to extract the maximum amount of information and to create an adequate dataset for training and evaluating models, as well as to apply data augmentation during model training to increase dataset size. By researching the available semantic segmentation algorithms, it was decided that the encoder-decoder architecture networks are suitable for the problem of landslide detection. In order to obtain quality of results, a series of experiments were conducted where different configurations of the encoder-decoder architecture were tested under the same conditions.

• Keywords: LiDAR; klizišta; semantička segmentacija; duboko učenje; konvolucijske neuronske mreže
• Keywords (english): LiDAR; landslides; semantic segmentation; deep learning; convolutional neural networks
• Language: croatian
• URN:NBN: urn:nbn:hr:168:256428
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-14 10:02:46