Klasifikacija biljnih vrsta iz slika biljaka temeljena na dubokom učenju

master's thesis

• Title (english): Classification of plant species from plant images based on deep learning
• Author: Marija Tišljar
• Mentor: Alan Jović (mentor)
• Committee member: Alan Jović (predsjednik povjerenstva)
• Committee member: Jan Šnajder (član povjerenstva)
• Committee member: Marko Čupić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-02-19, Croatia

Abstract

Učenje modela na skupu podataka sa velikom neuravnoteženosti klasa je izazovan problem. Ovaj rad kroz niz eksperimenata i analizu rezultata dolazi do zaključka da augmentacija podataka igra važnu ulogu u povećanju robusnosti i točnosti modela. Tehnika naduzorkovanja, također se pokazala efikasnom u rješavanju problema neuravnoteženosti klasa, doprinoseći boljem prepoznavanju manje zastupljenih klasa.
Odabrana arhitektura modela, ResNet50, u kombinaciji s augmentacijom i optimizacijom hiperparametara, pokazala se uspješnom u postizanju visoke točnosti klasifikacije, dosegavši klasifikacijsku točnost od približno 81 %, što je usporedivo sa srodnim rezultatima iz literature.
Dodatno, razvoj web aplikacije omogućio je interakciju korisnika s modelom u stvarnom vremenu.

Abstract (english)

Training models on a dataset with significant class imbalance is a challenging problem. This thesis, through a series of experiments and result analysis, concludes that data augmentation plays an important role in enhancing the robustness and accuracy of the model. The technique of oversampling has also proved effective in addressing the problem of class imbalance, contributing to better recognition of less represented classes.
The chosen model architecture, ResNet50, combined with data augmentation and hyperparameter optimization, has been successful in achieving high classification accuracy, reaching accuracy of 81 %, which is comparable to the results of related work.
Furthermore, the development of a web application has enabled real-time user interaction with the model.

• Keywords: augmentacija podataka; neuravnoteženost klasa; klasifikacija slika; duboko učenje; naduzorkovanje; neuronske mreže; PlantNet-300K; ResNet50
• Keywords (english): data augmentation; class imbalance; image classification; deep learning; oversampling; neural networks; PlantNet-300K; ResNet50
• Language: croatian
• URN:NBN: urn:nbn:hr:168:819419
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2024-03-28 07:55:37