Klasifikacija pokrovnosti z uporabo podatkov OpenStreetMap in satelitskih posnetkov Sentinel-2 : magistrsko delo

Jamnik, Uroš (author), Oštir, Krištof (mentor)

V magistrski nalogi sem predstavil postopek izdelave kart pokrovnosti z brezplačnimi in prosto dostopnimi podatki. Takšen postopek klasifikacije tal je ponovljiv, ne le za območje Slovenije ampak, za katero koli območje na Zemlji. Predstavljena sta pojma pokrovnosti in rabe tal. Opisane so najpomembnejše zbirke o rabi in pokrovnosti tal za območje Slovenije. Predstavljeni so načini klasifikacije, razvrščanje v razrede in izvajanje analize kakovosti klasifikacije. Kot vir satelitskih posnetkov za izdelavo naloge je predstavljen program Copernicus in satelita Sentinel-2. Opisan je projekt OpenStreetMap, ki sem ga uporabil za vir učnih vzorcev klasifikacije. Izdelal sem tri karte pokrovnosti za izbrano območje na osnovi učnih vzorcev zemljevida OpenStreetMap. Vir satelitskih posnetkov je satelit Sentinel-2. Karte se razlikujejo glede na število učnih vzorcev, saj je bil namen naloge ugotoviti optimalno število le teh za izdelavo karte pokrovnosti. Z analizo rezultatov sem ugotovil, da dva razreda pokrovnosti (razred gozd grmičevje in zaraščanje ter razred vode) dosegata zadovoljive rezultate. Za optimalno število učnih vzorcev pri izdelavi karte pokrovnosti se je izkazala karta s tremi učnimi vzorci. Domneve, ki sem jih preizkusil v magistrski nalogi sem potrdil s tem, ko sem pokazal, da je postopek klasifikacije mogoče opraviti samodejno, ter da zemljevid OpenStreetMap predstavlja dobre, vendar lahko tudi nezanesljive podatke.

• Language: Slovenian
• Keywords: pokrovnost, klasifikacija, OpenStreetMap, Sentinel-2, analiza natančnosti klasifikacije
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FGG - Faculty of Civil and Geodetic Engineering
• Publisher: [U. Jamnik]
• Year: 2017
• PID: 20.500.12556/RUL-96470
• UDC: 551.501.8:528.9(043.3)
• COBISS.SI-ID: 8207713
• Publication date in RUL: 02.10.2017

Secondary language

• Language: English
• Title: Land cover classification by using OpenStreetMap data and Sentinel-2 satellite imagery : Master thesis

Abstract

In my master’s thesis I presented the process of land cover maps production with free and publicly available data. Such land classification process is repeatable not only in the area of Slovenia but anywhere in the world. Concepts of land cover and land use are interpreted and the most important collections of land cover and use are chronicled. Various ways of classification, class categorization and performance of classification quality analysis are introduced as well. As the source of satellite imagery used the Copernicus program and Sentinel-2 satellites are represented. OpenStreetMap project, used as the source of learning samples, is also described. I made three land cover maps of chosen land based on learning samples of OpenStreetMap map. Sentinel-2 satellite is the source of satellite imagery. Maps differ according to number of learning samples on account of thesis' objective, which was to identify the optimal number of samples needed to produce a land cover map. By analyzing the results I determined the two classes of land cover (class of forest, bushes and overgrowth land and class of water) are reaching satisfying results. Assumptions tested in the master’s thesis were confirmed by showing that classification process can be carried out automatically. I also demonstrated that OpenStreetMap data displays valuable but occasionally unreliable data.

• Keywords: land cover, classification, OpenStreetMap, Sentinel-2, classification accuracy analysis