prikaz prve stranice dokumenta Klasifikacija slika metodama dubokog učenja
  Download
PDF 2.47 MB
master's thesis
Klasifikacija slika metodama dubokog učenja
2017. urn:nbn:hr:200:527518
Novoselnik, Filip
Josip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
Department of Computer Engineering and Automation
Chair of Automation and Robotics

Cite this document

Novoselnik, F. (2017). Klasifikacija slika metodama dubokog učenja (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:200:527518

Novoselnik, Filip. "Klasifikacija slika metodama dubokog učenja." Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2017. https://urn.nsk.hr/urn:nbn:hr:200:527518

Novoselnik, Filip. "Klasifikacija slika metodama dubokog učenja." Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2017. https://urn.nsk.hr/urn:nbn:hr:200:527518

Novoselnik, F. (2017). 'Klasifikacija slika metodama dubokog učenja', Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, accessed 19 November 2024, https://urn.nsk.hr/urn:nbn:hr:200:527518

Novoselnik F. Klasifikacija slika metodama dubokog učenja [Master's thesis]. Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek; 2017 [cited 2024 November 19] Available at: https://urn.nsk.hr/urn:nbn:hr:200:527518

F. Novoselnik, "Klasifikacija slika metodama dubokog učenja", Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, Osijek, 2017. Available at: https://urn.nsk.hr/urn:nbn:hr:200:527518

Please login to the repository to save this object to your list.
Metadata
TitleKlasifikacija slika metodama dubokog učenja
Title (english)Image classification using deep learning
AuthorFilip Novoselnik
MentorRatko Grbić (mentor)
Committee memberRatko Grbić (predsjednik povjerenstva)
Committee memberRobert Cupec (član povjerenstva)
Committee memberIrena Galić (član povjerenstva)
GranterJosip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
(Department of Computer Engineering and Automation)
(Chair of Automation and Robotics)
Osijek
Defense date and country2017-12-14, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Computing
Artificial Intelligence
Abstract
Područje dubokih neuronskih mreža doživljava nagli uzlet u posljednjih nekoliko godina te postiže izvrsne rezultate u području obrade slike i računalnog vida. Diferencijalna krvna slika, tj. analiza bijelih krvnih stanica je jedno od zanimljivih područja primjene ovakvih metoda budući da se trenutno radi vizualnom inspekcijom krvnih preparata što je vremenski zahtjevno ili vrlo skupom i sofisticiranom opremom. U sklopu ovog rada predložen je algoritam za detekciju i identifikaciju bijelih krvnih stanica na krvnim razmazima koji se dobivaju kamerom i mikroskopom. Prva faza obrade je detekcija bijelih krvnih stanica segmentacijom slike kojom se nastoje izdvojiti sve bijele krvne stanice. U drugoj fazi je treniran model konvolucijske neuronske mreže na temelju 168 označenih slika dobivenih iz Kliničkog Bolničkog Centra Osijek. Implementirani model klasificira stanice u 5 klasa: eozinofili, limfociti, monociti, segmentirani neutrofili i nepoznato, tj. slučaj kada je došlo do pogreške segmentacije. Točnost segmentacije je blizu 90 %, dok je točnost klasifikacije oko 81 %. Predloženi algoritam dao je vrlo dobre rezultate te demonstrirao potencijal korištenih metoda koji bi u potpunosti bio iskorišten kada bi model bio treniran na puno većem broju slika.
Abstract (english)
During the last few years, area of deep learning is growing and advancing rapidly, achieving very good results in image processing and computer vision. Differential blood count (white blood cells count) is one of the areas where such methods could be applied. Currently, it is done by visual inspection which is very time consuming or by using very expansive and sophisticated equipment. In this paper, a new system for white blood cells detection and identification is described and implemented. The first step is image segmentation, which extracts single cells out of original image. The second step is deep convolutional neural network training based on 168 labeled images from Clinical Hospital Center Osijek. Implemented model classifies white blood cells into five classes: eosinophiles, lymphocytes, monocytes, neutrphiles or unknown object (handles the case when segmentation has failed). Accuracy of segmentation on test samples is around 90 %, while accuracy of classification on test samples is around 81 %. Taking into account relatively small number of training samples implemented system showed very good results. It has demonstrated the potential of convolutional neural networks that would be fullfilled on much larger image dataset.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:200:527518
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 on2017-12-29 09:43:28