Fire Detection by Parallel Classification of Fire and Smoke Using Convolutional Neural Network

Fire detection is considered as a part of remote surveillance in domestic, industrial and the areas that are not approachable by human like deep forests. In this paper, convolutional neural network (CNN) is used to detect fire by classifying both fire and smoke in videos. A sequence of 2D convolutional layers and max pool layers is used to convert the video frames into feature maps with lower rank. The neural network is trained with the videos containing both fire and smoke. The videos with either fire or smoke or both are tested for fire detection with the FIRESENSE and other such open-source databases. The results show that the proposed method can classify the fire, smoke and fire with smoke with a recognition rate of up to 94%, 95% and 93%, respectively.