prikaz prve stranice dokumenta Primjena neuronskih mreža u klasificiranju LEGO kocaka
  Download
PDF 1.57 MB
master's thesis
Primjena neuronskih mreža u klasificiranju LEGO kocaka
2021. urn:nbn:hr:200:399760
Tufeković, Anto
Josip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
Department of Software Engineering
Chair of Visual Computing

Cite this document

Tufeković, A. (2021). Primjena neuronskih mreža u klasificiranju LEGO kocaka (Master's thesis). Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek. Retrieved from https://urn.nsk.hr/urn:nbn:hr:200:399760

Tufeković, Anto. "Primjena neuronskih mreža u klasificiranju LEGO kocaka." Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2021. https://urn.nsk.hr/urn:nbn:hr:200:399760

Tufeković, Anto. "Primjena neuronskih mreža u klasificiranju LEGO kocaka." Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2021. https://urn.nsk.hr/urn:nbn:hr:200:399760

Tufeković, A. (2021). 'Primjena neuronskih mreža u klasificiranju LEGO kocaka', Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, accessed 19 April 2024, https://urn.nsk.hr/urn:nbn:hr:200:399760

Tufeković A. Primjena neuronskih mreža u klasificiranju LEGO kocaka [Master's thesis]. Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek; 2021 [cited 2024 April 19] Available at: https://urn.nsk.hr/urn:nbn:hr:200:399760

A. Tufeković, "Primjena neuronskih mreža u klasificiranju LEGO kocaka", Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, Osijek, 2021. Available at: https://urn.nsk.hr/urn:nbn:hr:200:399760

Please login to the repository to save this object to your list.
Metadata
TitlePrimjena neuronskih mreža u klasificiranju LEGO kocaka
Title (english)LEGO Bricks Classification with Neural Networks
AuthorAnto Tufeković
MentorJosip Job (mentor)
Committee memberJosip Job (predsjednik povjerenstva)
Committee memberRatko Grbić (član povjerenstva)
Committee memberČaslav Livada (član povjerenstva)
GranterJosip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
(Department of Software Engineering)
(Chair of Visual Computing)
Osijek
Defense date and country2021-10-12, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Computing
Artificial Intelligence
Abstract
Korištenje dubokih umjetnih neuronskih mreža je postao de-facto standard kao prvi korak pri pokušavanju rješavanja nekog problema vezanog za prepoznavanje slika. Ovakvim načinom se dobije znatna količina znanja koju možemo upotrijebiti za analizu korištenih podataka i za daljnje korake potrebne za poboljšavanje korištene mreže (npr. promijene u proširenju skupa podataka ako je preveliko, tipa, smicanje). U ovom radu su opisane osnove o strojnom učenju i istraženo je kreiranje raznih mreža nad skupom podataka generiranih u Blender-u. Na kraju je odabrana duboka umjetna neuronska mreža temeljena na prijenosnom učenju od VGG16 mreže. Ta mreža je postigla najbolje rezultate (sitno više nego VGG19 temeljena mreža) i po njoj su istraženi ostali parametri kao što su matrica zabune i njene slike značajke, po prvoj metodi je određeno na kojim modelima mreža ima problema dok po drugoj metodi se otkriva kako mreža vidi dane slike u konvolucijskih blokovima.
Abstract (english)
The usage of artificial neural networks has become the de-facto standard as the first step in trying to solve a problem related to image recognition. This way one can obtain knowledge about the problem and use it for further analysis concerning the solution (i.e., introducing changes in the data augmentation steps to see how the deep network reacts). This work explains the basics of machine learning, and it explores the creation of various artificial neural networks which were then trained over an image data set generated in Blender. In the end, the deep artificial neural network based on transfer learning from VGG16 was chosen for its superior characteristics (although the VGG19 based one was close). The network was further explored for its properties, such as its confusion matrix and feature maps, the first research method reveals problematic classes which the network might fail to handle, and the second research method reveals how the network sees the given pictures as the pictures travel down the block of convolution layers.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:200:399760
Study programmeTitle: Graduate University Study Programme in Computer Engineering
Study programme type: university
Study level: graduate
Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
Type of resourceText
File originBorn digital
Access conditionsOpen access
Terms of useLicence In copyright icon
Created on2021-11-25 14:19:11