FIRE DETECTION BY TRANSFER LEARNING WITH EFFICIENTNET

Abstract

Fires often cause huge damage of wealth and lives. Traditional fire detection systems based on temperature and smoke sensors are able to detect fires when they are in dangerous stages and thus have limitations in extinguishing to minimize damages. This paper proposes a new method for early fire detection by applying transfer learning with EfficientNet-B3, a CNNs model. Images captured from camera are predicted to quickly discover if there is fire or not. If there is fire, then the alarm and fire extinguishing systems might be activated. Our system is implemented with Tensorflow and experimented upon public dataset. The obtained results show that its accuracy is high (97.5%) and can be applied in real life applications.