RT info:eu-repo/semantics/article
T1 Smoke detection in images through fractal dimension-based binary classification
A1 Pozo Velázquez, Javier del
A1 Aguiar Pérez, Javier Manuel
A1 Chamorro Posada, Pedro
A1 Pérez Juárez, María Ángeles
A1 Wang, Xinheng
A1 Casaseca de la Higuera, Juan Pablo
A1 Del-Pozo-Velázquez, Javier
A1 Aguiar-Pérez, Javier Manuel
A1 Chamorro-Posada, Pedro
A1 Pérez-Juárez, María Ángeles
A1 Casaseca-de-la-Higuera, Pablo
K1 Early fire detection
K1 Fractal dimension
K1 Image classification
K1 Remote Sensing
K1 33 Ciencias Tecnológicas
K1 1204 Geometría
AB Early fire detection is crucial for enabling rapid response and minimizing potentially catastrophic consequences. While artificial intelligence-based approaches have been developed for this task, they often demand substantial computational resources. Moreover, detecting smoke is inherently challenging due to its irregular, heterogeneous texture—especially under adverse weather conditions such as fog or cloud shadows. This paper introduces and validates an efficient smoke detection method grounded in fractal dimension analysis. The proposed approach involves dividing images into tiles, computing the fractal dimension for each block, and analysing the resulting fractal dimension distribution patterns to identify smoke presence. To evaluate its performance, we employed publicly available surveillance images from the High Performance Wireless Research and Education Network (HPWREN). Experimental results across five different scenarios demonstrate that the method achieves an accuracy of 96.87 %, successfully distinguishing between smoke and smoke-free regions—even under visually challenging conditions. By relying on an efficient fractal dimension algorithm, the proposed method is computationally efficient, and manages to capture the intrinsic texture characteristics of smoke, remaining unaffected by environmental noise such as fog and cloud cover.
PB Elsevier
SN 1051-2004
YR 2025
FD 2025
LK https://uvadoc.uva.es/handle/10324/76981
UL https://uvadoc.uva.es/handle/10324/76981
LA eng
NO Digital Signal Processing, 2025, vol. 166, p.105346
NO Producción Científica
DS UVaDOC
RD 01-ago-2025