Smoke detection in images through fractal dimension-based binary classification

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.