prikaz prve stranice dokumenta Semantička segmentacija slika prirodnog krajolika temeljena na dubokom učenju
Access restricted to students and staff of home institution
doctoral thesis
Semantička segmentacija slika prirodnog krajolika temeljena na dubokom učenju
Split: University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, 2020. urn:nbn:hr:179:179169
Božić-Štulić, Dunja
University of Split
Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture
Department of Electronics and Computing

Institutional repository: Repository of the Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture

Cite this document

Božić Štulić, D. (2020). Semantička segmentacija slika prirodnog krajolika temeljena na dubokom učenju (Doctoral thesis). Split: University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture. Retrieved from https://urn.nsk.hr/urn:nbn:hr:179:179169

Božić Štulić, Dunja. "Semantička segmentacija slika prirodnog krajolika temeljena na dubokom učenju." Doctoral thesis, University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, 2020. https://urn.nsk.hr/urn:nbn:hr:179:179169

Božić Štulić, Dunja. "Semantička segmentacija slika prirodnog krajolika temeljena na dubokom učenju." Doctoral thesis, University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, 2020. https://urn.nsk.hr/urn:nbn:hr:179:179169

Božić Štulić, D. (2020). 'Semantička segmentacija slika prirodnog krajolika temeljena na dubokom učenju', Doctoral thesis, University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, accessed 04 May 2024, https://urn.nsk.hr/urn:nbn:hr:179:179169

Božić Štulić D. Semantička segmentacija slika prirodnog krajolika temeljena na dubokom učenju [Doctoral thesis]. Split: University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture; 2020 [cited 2024 May 04] Available at: https://urn.nsk.hr/urn:nbn:hr:179:179169

D. Božić Štulić, "Semantička segmentacija slika prirodnog krajolika temeljena na dubokom učenju", Doctoral thesis, University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, Split, 2020. Available at: https://urn.nsk.hr/urn:nbn:hr:179:179169

Metadata
TitleSemantička segmentacija slika prirodnog krajolika temeljena na dubokom učenju
Title (english)Semantic segmentation of natural landscape images based on deep learning
AuthorDunja Božić-Štulić https://orcid.org/0000-0001-5267-0848
MentorDarko Stipaničev (mentor)
Committee memberVladan Papić (predsjednik povjerenstva)
Committee memberSven Lončarić (član povjerenstva)
Committee memberSaša Mladenović (član povjerenstva)
Committee memberDamir Krstinić (član povjerenstva)
Committee memberViktor Sučić (član povjerenstva)
GranterUniversity of Split
Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture
(Department of Electronics and Computing)
Split
Defense date and country2020-06-30, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Computing
Artificial Intelligence
Universal decimal classification
(UDC)		621.3 - Electrical engineering
Abstract
Semantička segmentacija slike značajni je zadatak u području računalnog vida i strojnog učenja. Ova doktorska disertacija bavi se razvojem nove metode za semantičku segmentaciju slika prirodnog krajolika. Modifikacijom bazne konvolucijske mreže predložen je novi model za semantičku segmentaciju slika. Pri izradi modela posebna pažnja je posvećena slojevima sažimanja, koji u modelima dubokog učenja predstavljaju veliki problem zbog gubitka podataka, koji je u kasnijim slojevima mreže nemoguće vratiti. Novi model ima segment posebno zadužen za rekonstrukciju izgubljenih podataka, pa je probleme gubitka podataka sveden na minimum u odnosu na postojeće mreže. Predložen je novi završni sloj koji poboljšava rekonstrukciju piksela slike izgubljenih u slojevima sažimanja. Evaluacija modela potvrdila je početne pretpostavke. Drugi dio ove disertacija je razvoj metode za proširenje referentne baze slika prirodnog krajolika i to posebno slika koje sadrže dim požara raslinja u početnim fazama gorenja. Predložen je i evaluiran model za generiranje realističkih sintetičkih slika dima požara raslinja primjenom generičkih suparničkih mreža. Ove su slike uspješno korištene u fazi treniranja mreže za semantičku klasifikaciju slika prirodnog krajolika i to posebno regija koje predstavljaju dim požara raslinja, što će značajno unaprijediti detekciju u automatskim sustavima za rano otkrivanje požara raslinja.
Abstract (english)
Semantic segmentation of regions on images is a significant task in the field of computer vision and machine learning. This doctoral dissertation deals with the development of a new method for the semantic segmentation of images of the natural landscape images. By modifying the base convolution network, a new model for semantic image segmentation is proposed. Particular attention was given to the compression layers, which in deep learning models pose a major data loss problem that is impossible to recover in later layers of the network. The new model has a segment specifically in charge of reconstructing lost data, so data loss problems are minimized over existing networks. A new final level is proposed that enhances the reconstruction of image pixels lost in the compression layers. Model evaluation confirmed the initial assumptions. The second part of this dissertation is the development of a method to extend the reference base of images of the natural landscape, especially images containing the wildfire smoke in the initial stages of burning. A model for generating realistic synthetic images of wildfire smoke using Generative Adversarial Network has been proposed and evaluated. These images have been successfully used in the training phase of the network for the semantic classification of the natural landscape images, especially the regions that represent wildfire smoke. This will probably significantly enhance detection in automated systems for early wildfire detection.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:179:179169
Promotion2022
Study programmeTitle: Electrical Engineering and Information Technology
Study programme type: university
Study level: postgraduate
Academic / professional title: doktor/doktorica znanosti, područje tehničkih znanosti, polje elektrotehnika (doktor/doktorica znanosti, područje tehničkih znanosti, polje elektrotehnika)
Type of resourceText
Extent125 str.
File originBorn digital
Access conditionsAccess restricted to students and staff of home institution
Terms of useLicence In copyright icon
Created on2022-02-28 10:47:07