Fire Identification Based on Novel Dense Generative Adversarial Networks

Abstract

Increasing death rates, damage to properties, and loss of trees can be caused by fires. In Australia and the United States of America, many fire incidents are reported annually. Due to that, both governments struggle from the devastation beyond plants, buildings, and infrastructure. A lot of people have lost their properties and land. Various innovations in fire detection technologies have been implemented to minimize the impacts of fires on the economy and lives. Some of these solutions are costly, while others lack accuracy. In this article, a novel deep-learning model to detect fires is presented. This model is based on new Novel Dense Generative Adversarial Networks (NDGANs) and image preprocessing technologies for fire detection through a continuous monitoring system. This system produces alarms if a fire or smoke is detected. The proposed approach was trained and tested on five datasets. This system was evaluated using four performance quantities, which are accuracy, sensitivity, dice, and F-score, and attained 98.87%, 97.64%, 98.82%, and 98.69% for the considered quantities, respectively. In addition, the proposed method was compared with other developed approaches and outperformed these methods. The presented New Dense Generative Adversarial Networks technology is useful in fire detection as shown from the conducted simulation experiments on MATLAB.