Title Klasifikacija slika primjenom nadziranog strojnog učenja Title (english) Image classification using supervised machine learning Author Karla Roginić Mentor Sanja Seljan (mentor)Committee member Ivan Dunđer (predsjednik povjerenstva)Committee member Vjera Lopina (član povjerenstva)Committee member Sanja Seljan (član povjerenstva)Granter University of Zagreb Defense date and country 2024-09-20, Croatia Scientific / art field, SOCIAL SCIENCES Abstract Prisutnost tehnologije u svakom aspektu života postala je gotovo neizostavna. Tehnološki napredak transformira način na koji živimo i utječe na naše razumijevanje i interpretaciju svijeta oko nas. Posebno se očituje napredak u području strojnog učenja koje je zahvatilo sve slojeve znanosti i većinu industrija, posebno modnu industriju. Stoga primjena nadziranog strojnog
učenja u klasifikaciji slika s naglaskom na kulturno specifične haljine iz različitih regija svijeta postaje važna. Ovim pristupom želi se doprinijeti boljem razumijevanju i kategorizaciji kulturno specifičnih odjevnih tradicija, s potencijalnim primjenama u modnoj industriji, obrazovanju i promicanju kulturne raznolikosti. Cilj ovog rada bio je razviti model za klasifikaciju slika koji može prepoznati i klasificirati raznovrsne haljine prema njihovom podrijetlu. Razvijen je model za klasifikaciju slika koristeći
skup podataka koji sadrži slike kulturno raznolikih haljina iz Skandinavije, Mediterana, Ujedinjenih Arapskih Emirata i središnje Europe. Model je treniran na različitom broju epoha (50 i 100) kako bi se usporedili rezultati i performanse modela s obzirom na broj epoha. Analiza performansi modela u odnosu na broj epoha, pokazuje da model treniran na većem broju epoha
(100) postiže neznatna poboljšanja u odnosu na model treniran na manjem broju epoha (50). Veći broj epoha (100) poboljšava performanse u klasifikaciji slika haljina karakterističnih za Ujedinjene Arapske Emirate i Skandinaviju, dok negativno utječe na klasifikaciju slika haljina karakterističnih za središnju Europu i Mediteran. I model treniran na 50 epoha i model treniran na 100 epoha ima poteškoća u klasificiranju slika haljina karakterističnih za središnju Europu, dok najbolje rezultate postiže u klasifikaciji slika haljina karakterističnih za Mediteran. Za postizanje optimalnih rezultata potrebno je daljnje podešavanje broja epoha i drugih hiperparametara. Također, povećanje broja i raznolikosti skupa podataka, s ciljem boljeg
odražavanja različitih uvjeta i karakteristika haljina, može značajno unaprijediti sposobnost modela da točno klasificira slike u sve kategorije.
Abstract (english) The presence of technology in every aspect of life has become almost inevitable. Technological progress is transforming the way we live and influencing our understanding and interpretation of the world around us. The progress in the field of machine learning, which has affected all layers of science and most industries, especially the fashion industry, is particularly evident. Therefore, the application of supervised machine learning in image classification with a focus on culturally specific dresses from different regions of the world becomes important. This approach aims to contribute to a better understanding and categorization of culturally specific clothing traditions, with potential applications in the fashion industry, education and the promotion of cultural diversity. The aim of this paper was to develop a model for image classification that can recognize and classify various dresses according to their origin. An image classification model was developed, using a dataset containing images of culturally diverse dresses from Scandinavia, the Mediterranean, the United Arab Emirates, and Central Europe. The model was trained on different numbers of epochs (50 and 100) to compare the results and performance of the model with respect to the number of epochs. The analysis of the performance of the model in relation to the number of epochs shows that the model trained on a larger number of epochs (100) achieves slight improvements compared to the model trained on a smaller number of epochs (50). A larger number of epochs (100) improves the performance in the classification of dress images characteristic of the United Arab Emirates and Scandinavia, while negatively affectting the classification of dress images characteristic of Central Europe and the Mediterranean. Both the model trained on 50 epochs and the model trained on 100 epochs have difficulties in classifying images of dresses characteristic of Central Europe, while they achieve the best results in classification of images of dresses characteristic of the Mediterranean. To achieve optimal results, further adjustment of the number of epochs and other hyperparameters is necessary. Also, increasing the number and diversity of the dataset, with the aim of better
reflecting the different conditions and characteristics of the dresses, can significantly improve the model's ability to accurately classify images into all categories.
Keywords
strojno učenje
nadzirano strojno učenje
klasifikacija slika
detekcija objekata
modna industrija
epohe
Keywords (english)
machine learning
supervised machine learning
image classification
object detection
fashion industry
epochs
Language croatian URN:NBN urn:nbn:hr:131:076804 Study programme Title: Information Sciences; specializations in: Archivistics Study, Library Science, Information Sciences (research), Information Sciences (teaching), Museology and heritage management, Informatology Type of resource Text File origin Born digital Access conditions Open access Terms of use Created on 2024-09-20 13:16:26
