Publication

Abstract : A key element in wildfire combat is early and efficient detection. Early warning activities centered on early intervention, specific results both during the day and night, and the ability to prioritize the risk of fire. Vision-based fire detection has recently gained popularity over the classic sensor-based fire detection systems. However, the process of identification by the technique of image processing is repetitive. The reason to use deep learning is that these models can create features without a human intervention. The performance of five different models for forest fire classification is analyzed in this paper: VGG-16, ResNet-50-V2, MobileNet-V2, Inception-V2, and Xception. MobileNet-V2 performed the best among all the architectures with an accuracy of 96.84% on the dataset.