Analiza vegetacijskog pokrova na području Varaždinske županije metodama strojnoga učenja

master's thesis

• Title (english): The analysis of the vegetation cover in the area of Varaždin County by using machine learning methods
• Author: Magdalena Remetović
• Mentor: Nikola Kranjčić (mentor)
• Committee member: Sanja Kovač (predsjednik povjerenstva)
• Committee member: Nikola Kranjčić (član povjerenstva)
• Committee member: Davor Stanko (član povjerenstva)
• Committee member: Jasmin Jug (član povjerenstva)
• Granter: University of Zagreb; Faculty of Geotechnical Engineering; Varaždin
• Defense date and country: 2022-09-19, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Interdisciplinary Technical Sciences; Environmental Engineering

Abstract

Cilj ovog istraživanja je analiza vegetacijskog pokrova Varaždinske županije upotrebom
metoda strojnoga učenja. Izdvojene su najčešće korištene metode strojnoga učenja kod analize
vegetacijskog pokrova kroz raspoznavanje uzoraka i dubinsku analizu podataka. Rad daje
pregled značajnih i najvažnijih satelitskih misija za snimanje Zemljine površine koje
omogućuju prikupljanje i upotrebu satelitskih snimaka visoke i vrlo visoke rezolucije. U radu
su korištene Sentinel-2 snimke područja Varaždinske županije. Analizirana su u 4 različita
vremenska razdoblja između 2015. i 2021. godine, a za obradu satelitskih snimaka, klasifikaciju
i naknadne analize korišteni su programi otvorenoga koda SAGA-GIS i QGIS. Na kraju je
provedena analiza vegetacijskog pokrova pomoću vegetacijskog indeksa NDVI nakon čega su
prikazani rezultati i doneseni zaključci o promjenama vegetacijskog pokrova na promatranom
području. Rezultati metode potpornih vektora ukazuje da se vegetacijski pokrov na području
Varaždinske županije smanjio za 12,94% u roku od 6 godina. Metoda slučajne šume pokazuje
slične rezultate sa smanjenjem od 10,55%, dok klasifikator maksimalne vjerojatnosti pokazuje
smanjenje za 8,42%. Rezultati NDVI pokazuju smanjenje vrlo visoke pokrivenosti vegetacijom
za 10%. Zaključak ovog rada je da su veći uzorci uzeti za šume, nisku vegetaciju i vode
omogućili bolju klasifikaciju dok su manji uzorci oranica i izgrađenog područja dali lošiju
klasifikaciju tih klasa na promatranom području. Najboljom metodm pokazao se klasifikator
maksimalne vjerojatnosti s najtočnijim i najpreciznijim rezultatima klasifikacije svih klasa, a
sve metode pokazale su se vrlo dobre za analizu vegetacijskog pokrova na promatranom
području.

Abstract (english)

The aim of this research is to analyse the vegetation cover of Varaždin County using machine
learning methods. The most commonly machine learning methods used in the analysis of
vegetation cover through pattern recognition and in - depth data analysis were singled out. The
paper provides an overview of the significant and most important satellite missions for imaging
of the Earth 's surface that have enabled the collection and use of high and very high resolution
satellite imagery. Sentinel-2 images of Varaždin County area were used in this paper. They
were analysed in 4 different time periods between 2015 and 2021, and for the processing of
satellite images, their classification and subsequent analyses open source programs SAGA-GIS
and QGIS were used. Finally, the analysis of vegetation cover was performed using the
vegetation index NDVI, after which the results were presented and a conclusion was reached
on changes in vegetation cover in the observed area. The results of the support vector machines
show that the vegetation cover in Varaždin County area has decreased by 12.94% within 6
years. A random forest shows similar results with a reduction of 10.55%, while the maximum
likelihood classifier shows a reduction of 8.42%. The NDVI results show a 10% reduction in
very high vegetation cover. The conclusion of this paper is that larger samples taken for forests,
low vegetation and water allowed better classification, while smaller samples of arable land and
built-up areas gave worse classification of these classes in the observed area. The best method
was the maximum likelihood classifier with the most accurate and precise classification results
for all classes, and all the methods proved to be very good for analysing the vegetation cover in
the observed area.

• Keywords: metode strojnoga učenja; daljinska istraživanja; vegetacijski indeks; QGIS
• Keywords (english): machine learning methods; remote sensing; vegetation index; QGIS
• Language: croatian
• URN:NBN: urn:nbn:hr:130:167637
• Study programme: Title: Geoenvironmental Engineering; specializations in: Geoenvironmental Engineering, Environmental Management, Water Management; Course: Geoenvironmental Engineering; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/ inženjerka inženjerstva okoliša (magistar/magistra inženjer/ inženjerka inženjerstva okoliša)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Access restricted to authenticated users
• Created on: 2022-12-12 09:58:06