Umjetne neuronske mreže u prepoznavanju teksta

undergraduate thesis

• Title (english): Artificial neural networks in text recognition
• Author: Mario Branimir Fraćin
• Mentor: Dubravko Majetić (mentor)
• Granter: University of Zagreb; Faculty of Mechanical Engineering and Naval Architecture; Zagreb
• Defense date and country: 2019-09-26, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Mechanical Engineering

Abstract

Tema ovog završnog rada je upotreba umjetnih neuronskih mreža pri prepoznavanju rukom pisanih brojeva i teksta. Uvodno poglavlje daje uvid u sam problem, te ukratko opisuje strukturu i podjelu umjetnih neuronskih mreža. U drugom poglavlju je sam problem prepoznavanja brojki i slova detaljnije opisan, i uz njega su prikazana područja primjena u kojima se danas koriste takvi algoritmi. Zatim je provedena analiza najčešće korištenih algoritama u opisanom području istraživanja. Algoritmi dubokog učenja korišteni u programskoj podršci ovog rada su detaljnije obrađeni u trećem poglavlju, a za ostale suvremene algoritme je dan pregled visoke razine te je ukratko opisan njihov način rada. Konačno, u zadnjem poglavlju su izneseni rezultati rada različitih struktura umjetnih neuronskih mreža za rješavanje navedenog problema koristeći varijacije značajnih parametara. Modeli spomenutih struktura kreirani su u programskom paketu Python koristeći biblioteku TensorFlow uz razne dodatne biblioteke.

Abstract (english)

The topic of this thesis is the application of artificial neural networks in handwritten character and digit recognition. The problem is presented in the introductory chapter, along with types and structures of artificial neural networks. Second chapter gives a deeper insight to the problem of digit and character recognition, and describes contemporary applications of such technologies. The analysis of most commonly used algorithms in the described field of research is given afterwards. Deep learning methods used in this thesis are thoroughly analyzed in the third chapter, along with a high level overview of other contemporary algorithms and their operating principles. Finally, the last chapter contains the results of different structures of neural networks for the given problem with variations to the most important parameters. Models of previously mentioned structures are created in the programming language Python combined with a machine learning library TensorFlow, along with various additional libraries.

• Keywords: Umjetna inteligencija; strojno učenje; neuronske mreže; prepoznavanje brojki i teksta; digitalizacija teksta; obrada slike
• Keywords (english): Artificial intelligence; machine learning; neural networks; digit and text recognition; text digitalisation; image processing
• Language: croatian
• URN:NBN: urn:nbn:hr:235:518061
• Study programme: Title: Mechanical Engineering; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka strojarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka strojarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2019-09-23 09:09:40