Ontological model for fire image segmentation using an adaptive neuro-fuzzy network

Fire image segmentation is a critical task for early detection and emergency monitoring systems. Traditional methods, such as thresholding and complex computational models, face limitations in terms of speed and accuracy under realtime conditions. The objective of this study is to enable an image segmentation process operating in near real-time. This paper presents an ontological model for fire image segmentation based on an adaptive neuro-fuzzy system with an optimized defuzzifier employing the area ratio method. The proposed model structures data processing workflows, including image transformation into the hue-saturation-brightness color space, the use of triangular membership functions, and adaptive parameter tuning for identifying fire regions. The ontology systematizes classes, processes, and their interrelationships, thereby ensuring modularity and flexibility of the solution. Experiments on images of nighttime fires with smoke showed that the proposed model achieves high accuracy and speed, outperforming the compared segmentation methods, and demonstrates potential for application in real-time monitoring systems.