Early Forest Fire Region Segmentation Based on Deep Learning

As the forest fire can bring about great property loss and ecological disaster, artificial intelligence-based forest fire monitoring system has gained popularity in recent years to enable the fire alarm quickly and accurately. In this paper, considering that the fire area is very small and hard to be detected using traditional method for detection early forest fire, we propose a novel forest fire monitoring framework based on convolutional neutral networks. In order to validate that the proposed framework can improve effectiveness and accuracy of detecting the early forest fires, many groups of fire detection experiments using a self-generated forest fire dataset and two real forest fire monitor videos are conducted. The experiment results demonstrate its capability to work in various challenging fire and illumination conditions presented in the study, and show that the framework can effectively detect the early forest fire.