Supervisor

Sam Weiss

Programme

MSc in Data Analytics

Subject

Computer Science

Abstract

Plant disease detection is a critical challenge in agriculture, where limited annotated datasets often hinder the training of robust deep learning models. This thesis explores the application of generative adversarial networks for data augmentation in the classification of six classes of sorghum diseases, addressing the scarcity of real training data. A lightweight convolutional neural network  (CNN) was employed as the classifier, trained under three experimental conditions: using only real images, a stratified mix of real and DCGAN-generated samples, and a stratified mix of real and WGAN-GP-generated samples.

The baseline CNN, trained on 279 real images, achieved an accuracy of 89.7%, demonstrating strong performance given the limited dataset size. Incorporating DCGAN-generated images improved accuracy to 93.3%, with enhanced intra-class variability and reduced overfitting. By contrast, augmentation with WGAN-GP samples reduced accuracy to 86.1%, reflecting unstable training dynamics and inconsistent sample quality. Quantitative evaluation using Fréchet Inception Distance (FID) confirmed that both GANs struggled to generate high quality outputs, with DCGAN achieving slightly lower score. Furthermore, t-SNE shower cleared separation between real and synthetic feature distributions, indicating limited overlap and highlighting the challenges of realistic data generation at low resolution.

Overall, the study demonstrates that GAN-based augmentation can enhance classification performance under extreme data scarcity, though success depends strongly on generative stability image fidelity m and classifier robustness. DCGAN provided measurable gains , whereas WGAN-GP underperformed due to insufficient hyperparameter tuning and computational restraints. The results emphasize the potential of GANs in agricultural imaging while underscoring the importance of scalable datasets, higher resolution models and advanced GAN architectures in future research.

Date of Award

2025

Full Publication Date

2025

Access Rights

open access

Document Type

Capstone Project

Resource Type

thesis

Included in

Data Science Commons

Share

COinS