Synthetic data generation with denoising diffusion probabilistic models for data augmentation in data-limited satellite image classification

Publication date

2025-12-01T15:03:27Z

2025-12-01T15:03:27Z

2025-06



Abstract

Treball fi de màster de: Master's Degree in Data Science. Methodology Program. Curs 2024-2025


Tutor: Antonio Lozano


Data augmentation is essential for improving deep learning performance with limited data. This thesis examines whether class-conditional Denoising Diffusion Probabilistic Models (DDPMs) can enhance satellite image classification on the EuroSAT dataset. Using a U-Net-based DDPM, we generated synthetic images for ten land cover classes and evaluated ResNet-18 with different real-to-synthetic ratios. Results show that geometric transformations consistently outperform synthetic data, which often degrades performance, especially at higher proportions. However, hybrid approaches improved specific classes, such as AnnualCrop (+2.65 points). Overall, geometric augmentation remains most effective, though class-dependent synthetic strategies show potential for targeted enhancement.


L’augmentació de dades és essencial per millorar el rendiment de l’aprenentatge profund quan les dades són limitades. Aquesta tesi analitza si els Denoising Diffusion Probabilistic Models (DDPM) condicionals per classe poden millorar la classificació d’imatges satel·litals al conjunt de dades EuroSAT. Mitjançant un DDPM basat en U-Net es van generar imatges sintètiques per a deu classes de coberta terrestre i es va avaluar ResNet-18 amb diferents proporcions de dades reals i sintètiques. Els resultats mostren que les transformacions geomètriques superen sistemàticament les dades sintètiques, tot i que els enfocaments híbrids van millorar classes específiques com AnnualCrop (+2.65 punts).

Document Type

Master's final project

Language

English

Recommended citation

This citation was generated automatically.

Rights

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License

https://creativecommons.org/licenses/by-nc-nd/4.0

This item appears in the following Collection(s)