Izvorni znanstveni članak
https://doi.org/10.24138/jcomss-2025-0245
Generation of Synthetic Wildfire Smoke Images with Generative Adversarial Networks
Dunja Božić-Štulić
; University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, Split, Croatia
Damir Krstinić
orcid.org/0000-0002-9946-6630
; University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, Split, Croatia
*
Darko Stipaničev
; University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, Split, Croatia
Jakov Bejo
orcid.org/0009-0000-1367-6630
; University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, Split, Croatia
* Dopisni autor.
Sažetak
The scarcity of annotated images significantly hinders
the development of robust deep learning models for early
wildfire s moke d etection. T raditional a ugmentation methods,
such as rotation or mirroring, are often insufficient. T his is
particularly true for detecting subtle smoke formations at long
distances, where smoke is often barely perceptible even to human
observers. Existing smoke datasets predominantly feature developed
smoke plumes or closer views, making them unsuitable for
training models for this critical early-phase detection. To address
this, we propose generating synthetic images using Generative
Adversarial Networks. Unlike typical GAN applications that aim
for high-fidelity o bject r epresentation, o ur o bjective i s different.
We synthesize realistic, fuzzy images of subtle, distant smoke
— blurred and blended with the background — yet retaining
characteristic features essential for classifier training. We propose
a GAN architecture based on a modified Super-Resolution GAN,
specifically a dapted w ithout B r esidual b locks, i n o rder to
produce realistic images of smoke at long distances. Experimental
evaluation demonstrates that augmenting datasets with GANgenerated
smoke images significantly improves t he performance
of classifiers in detecting early-stage wildfire smoke, affirming the
utility of GANs for data enhancement even when generating lowquality,
realistic imagery. This method mitigates data scarcity,
offering a viable solution for training effective early wildfire
detection systems.
Ključne riječi
generative adversarial networks; synthetic smoke images; deep learning
Hrčak ID:
348537
URI
Datum izdavanja:
31.3.2026.
Posjeta: 0 *