A Comparison Study for applying Different Deep Learning Models On Fire Detection method

Fire is one of the most dangerous types of accidents that can occur and may cause huge harm to people's lives and threaten an environment resources by very fast spreading, which makes it out of control very quickly, this important aspect requires an intelligent system to control it, the ability of fire detection system needs improvement all time. Convolutional Neural Networks (CNNs) have been proposed for many different high-performance computer vision missions. Although many CNN models nowadays have been proposed in this field and discussed, it is hard to conclude the best model to detect fire in terms of accuracy. This paper presents different CNN models (Vgg16, Vgg19, ResNet50, InceptionV3, and MobileNetV2) for this purpose. The performance of these models has been studied and compared with other CNN models introduced in the literature review section. Different preprocessing and augmentation techniques have been used for the input dataset (Fire and Non-fire images), and then train the proposed models on a new dataset that has been created. The results obtained in this paper revealed that MobileNet2 has outperformed the other CNN models with an overall accuracy up to 99.7%.