Automatizirano ispravljanje slika iz dronova temeljeno na dubokom učenju

master's thesis

• Title (english): Deep Learning for Automatized Correction of Images Taken from Drones
• Author: Bruno Horvat
• Mentor: Vladimir Čeperić (mentor)
• Committee member: Vladimir Čeperić (predsjednik povjerenstva)
• Committee member: Zoran Kalafatić (član povjerenstva)
• Committee member: Tvrtko Mandić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2023-07-13, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Ovim radom isprobane su razne metode učenja rekonstruiranja slike pomoću neuronske mreže. Najprije je dana teorijska podloga raznih vrsta modela te je opisano kako rade i koje su njihove prednosti. Opisane metode su služile kao kratak uvod u područje dubokog učenja. Bez prevelikog ulaženja u detalje, opisane su one metode koje će u ovom radu biti i korištene.
Za treniranje neuronske mreže je potreban skup podataka koji je raznim augmentacijama pokušao simulirati nepovoljne vremenske uvjete. Problem čišćenja slika je prikazan kroz analitičko razmišljanje te je napravljena usporedba modela i njihovih rezultata. Svaki model sastoji se od enkoderske i dekoderske mreže koje zajedničkim snagama pokušavaju rekonstruirati sliku iz drona. Modeli dubokog učenja naučeni su najviše što su mogli biti s obzirom na hardverske resurse.

Abstract (english)

The topic of the following paper is testing various methods of image reconstruction using a neural network. Firstly, a theoretical background of different types of models is provided, describing how they work and what are their advantages. The described methods served as a brief introduction to the field of deep learning. Without going into too much details, only the methods that are used in this work were described.
To train the neural network, a dataset was required, which was subjected to various augmentations in order to simulate weather conditions. The problem of image denoising has been shown through analytical thinking, and a comparison of models and their results was conducted. Each model consists of an encoder and a decoder network, which work together to reconstruct images captured by a drone. The deep learning models were trained to the best of their ability given the available hardware resources.

• Keywords: Maskirani autoenkoder; Konvolucijska neuronska mreža; Autoenkoderska arhitektura; Dron; Čišćenje slike
• Keywords (english): Masked autoencoder; Convolutional neural network; Autoencoder architecture; Drone; Image denoising
• Language: croatian
• URN:NBN: urn:nbn:hr:168:043951
• 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: 2023-10-11 09:02:20