Analiza učinkovitosti automatske segmentacije satelitskih snimki korištenjem Segment Anything Model-a i harmonijske analize NDVI vrijednosti

master's thesis

• Title (english): Analysis of the Effectiveness of Automatic Satellite Image Segmentation Using the Segment Anything Model and Harmonic Analysis of NDVI Values
• Author: Filip Papić
• Mentor: Mario Miler (mentor)
• Committee member: Mario Miler (predsjednik povjerenstva)
• Committee member: Damir Medak (član povjerenstva)
• Committee member: Luka Rumora (član povjerenstva)
• Granter: University of Zagreb; Faculty of Geodesy; (Institute of Geomatics); (Chair of Geoinformatics); Zagreb
• Defense date and country: 2024-09-06, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Geodesy; Geomatics

Abstract

Ovaj diplomski rad istražuje učinkovitost automatske segmentacije satelitskih snimki korištenjem Segment Anything Model-a (SAM) i harmonijske analize NDVI vrijednosti. Korištenjem satelitskih snimki područja Varaždina prikupljenih od strane PlanetScope satelita, provedena je harmonijska analiza NDVI vrijednosti koja omogućuje detaljniju analizu sezonskih promjena vegetacije. Izdvojene harmonijske komponente (amplituda, faza i srednja vrijednost) korištene su za rekoloriranje satelitskih snimki u HSL modelu boja, što je poboljšalo vizualnu interpretaciju snimki i rezultate segmentacije. Segment Anything Model primijenjen je na originalne i rekolorirane snimke, a rezultati su uspoređeni s podacima iz ARKOD sustava. Analiza je pokazala visoku razinu sukladnosti između rezultata segmentacije i ARKOD podataka, posebno za veće poljoprivredne čestice, dok su uočena manja odstupanja na manjim česticama. Korištenje HSL rekoloriranja omogućilo je bolje razlikovanje različitih tipova vegetacije i zemljišnih površina. U budućim istraživanjima preporuča se daljnja analiza potencijala HSL rekoloriranja u identifikaciji specifičnih kultura na temelju boja nakon rekoloriranja.

Abstract (english)

This thesis explores the effectiveness of automatic satellite image segmentation using the Segment Anything Model (SAM) and harmonic analysis of NDVI values. Using satellite images of the Varaždin area collected by PlanetScope satellites, a harmonic analysis of NDVI values was conducted, allowing for a detailed examination of seasonal vegetation changes. Extracted harmonic components (amplitude, phase, and mean value) were used to recolor satellite images in the HSL color model, which improved the visual interpretation of the images and segmentation results. The Segment Anything Model was applied to both original and recolored images, and the results were compared with data from the ARKOD system. The analysis revealed a high level of consistency between segmentation results and ARKOD data, particularly for larger agricultural plots, while minor discrepancies were noted in smaller plots. The use of HSL recoloring enabled better differentiation between various types of vegetation and land surfaces. Future research is recommended to further analyze the potential of HSL recoloring in identifying specific crops based on the colors observed post-recoloring.

• Keywords: harmonijska analiza; NDVI; rekoloriranje; segmentacija; satelitske snimke.
• Keywords (english): harmonic analysis; NDVI; recoloring; satellite images; segmentation.
• Language: croatian
• URN:NBN: urn:nbn:hr:256:560368
• Study programme: Title: Geodesy and Geoinformatics; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka geodezije i geoinformatike (magistar/magistra inženjer/inženjerka geodezije i geoinformatike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2024-09-02 17:19:40