Wildfire Flame and Smoke Detection Using Static Image Features and Artificial Neural Network

If forest fires are not contained quickly, they can spread wide very fast and cause devastating environmental, social and economic damages. The best method to minimize wildfire loss is to be able to detect it in its early stages for rapid containment and suppression. Fire comes with some distinguishable signatures such as flame, smoke and heat that can be used for early detection using computer vision based remote sensing techniques. Each signature has its own merits and demerits that vary under different environmental conditions and circumstances. Therefore, it is not always enough to form a detection algorithm based on a single signature. Keeping that in mind, this paper presents a novel algorithm that is capable of detecting both flame and smoke from a single image using block-based color features, texture features and a single artificial neural network (ANN). Such an algorithm is capable of providing reliable, rapid and continuous detection under any circumstances and can be incorporated into the existing unmanned aerial vehicle (UAV) based fire monitoring system.