Enhancing Plant Disease Classification through Unsupervised Domain Adaptation (UDA)

Nadigottu, Raj (2025) Enhancing Plant Disease Classification through Unsupervised Domain Adaptation (UDA). Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (729kB) | Preview
Preview	PDF (Configuration Manual) Download (391kB) | Preview

Abstract

Classification models of plant diseases have been proven to show significant accuracy when trained and tested within a well-controlled laboratory environment but show poor performance when tested on real farming environments, where there is a domain mismatch between clean lab and field images. In this paper, we examine how Unsupervised Domain Adaptation (UDA) methods may be used to overcome this gap in order to achieve generalization with no need of labelled target-domain data. With the PlantVillage dataset, as the source domain and PlantDoc as the target domain, the proposed pipeline combines three components: CycleGAN used to do the visual translation of the domain, SimCLR that learns feature representations in a self-supervised fashion, and pseudo-labelling, which uses a back-and-forth arrangement to incorporate the target-domain samples with high confidence.

MobileNetV2 classifier has been trained on the original and adapted datasets, and then tested it on a separate set of PlantDoc images. Some of the UDA approaches produced better results than the baseline, but the improvements were uneven. In the final evaluation, the model achieved 23.08% accuracy and a macro F1 score of 0.1628. These outcomes show that, while UDA offers potential benefits for agricultural image classification, it also has clear limitations. Factors such as uneven class distribution, the complexity of the datasets, and the presence of noisy labels appeared to play a role in restricting performance.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Thomas, Lavish UNSPECIFIED
Subjects:	S Agriculture > S Agriculture (General) S Agriculture > SB Plant culture Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence
Divisions:	School of Computing > Master of Science in Artificial Intelligence
Depositing User:	Ciara O'Brien
Date Deposited:	12 Aug 2026 09:21
Last Modified:	12 Aug 2026 09:21
URI:	https://norma.ncirl.ie/id/eprint/9510

Actions (login required)

View Item