Proučavanje asteroida u pregledima neba primjenom neuronskih mreža

master's thesis

• Title (english): Study of asteroids in sky surveys using neural networks
• Author: Tomislav Smolčić
• Mentor: Goranka Bilalbegović (mentor)
• Committee member: Goranka Bilalbegović (predsjednik povjerenstva)
• Committee member: Krešimir Kumerički (član povjerenstva)
• Committee member: Mihael Makek (član povjerenstva)
• Committee member: Davor Horvatić (član povjerenstva)
• Committee member: Mario Basletić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Science; (Department of Physics); Zagreb
• Defense date and country: 2018-07-12, Croatia
• Scientific / art field, discipline and subdiscipline: NATURAL SCIENCES; Physics

Abstract

Asteroidi pružaju neusporediv pogled u prošlost Sunčevog sustava, ali i predstavljaju potencijalnu opasnost za planet Zemlju. Detekcija i identifikacija asteroida od iznimne su važnosti. Zbog toga su i danas u tijeku astronomska istraživanja čiji je cilj otkrivanje novih i praćenje orbita već poznatih, ali potencijalno opasnih asteroida. U takvim istraživanjima generiraju se ogromne količine podataka zbog čega je važan razvoj računalnih metoda za njihovu brzu i točnu obradu. U ovom radu predlažemo neuronske mreže, algoritme strojnog učenja, za detekciju asteroida na fotografijama iz Sloanovog digitalnog pregleda neba (SDSS, engl. Sloan Digital Sky Survey) kao mogući smjer u razvoju računalnih sustava za obradu podataka dobivenih u astronomskim istraživanjima. Uspoređujemo učinak klasične i konvolucijske neuronske mreže koje su izvedene u programskom jeziku Python i njegovim paketima za strojno učenje. U Dodatku, kao metodički dio diplomskog rada, predstavljena je uporaba programskog jezika Python za obradu slika kao nastavna jedinica iz predmeta Informatika u prirodoslovno-matematičkim gimnazijama.

Abstract (english)

Asteroids give a specific insight into history of the Solar system, but also present a potential danger for the planet Earth. It is very important to detect and identify asteroids. Because of that, there are ongoing research projects with the purpose of finding new asteroids and tracking those already known, but potentially dangerous. In this research large amounts of data are generated. Therefore it is important to develop new computer methods to process data in the extremely fast and accurate way. In this thesis we propose neural networks, machine learning algorithms, for detecting asteroids in data from the Sloan Digital Sky Survey (SDSS). This is a possible direction for development of computational methods to process data gathered in astronomy. We compare performance of standard and a convolutional neural networks implemented in Python programming language and its machine learning libraries. In the Appendix, image processing in Python is presented as a topic for high school computer science classes.

• Keywords: asteroidi; strojno učenje; neuronske mreže; programski jezik Python; Scikit-Learn; TensorFlow
• Keywords (english): asteroids; machine learning; neural networks; Python programming language; SciKit-Learn; TensorFlow
• Language: croatian
• URN:NBN: urn:nbn:hr:217:997321
• Study programme: Title: Physics and Computer Science; specializations in: Education; Course: Education; Study programme type: university; Study level: integrated undergraduate and graduate; Academic / professional title: magistar/magistra fizike i informatike (magistar/magistra fizike i informatike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access; Embargo expiration date: 2020-07-12
• Created on: 2018-07-17 12:29:27