Fire Detection using Transfer Learning and Pre-Trained Model

Abstract

In recent years, there has been a rise in the occurrence of fire outbreaks, posing a significant threat to both natural resources and human lives. Consequently, the significance of fire detection in safeguarding human life and property has become increasingly crucial. Various studies have employed sensors and machine learning algorithms for fire detection, with CNN emerging as one of the most promising approaches in this field. The aim of this research paper is to explore the efficacy of using pre-trained CNN models, including VGG16, VGG19, MobileNet, and InceptionV3, through transfer learning to enhance the accuracy of the fire detection classification task. The models were trained on a dataset of 600 images and tested on 250 images, collected from various online sources and sorted into Fire and Non-Fire categories. Our findings indicate that the use of transfer learning in fire detection task yields higher accuracy than the base model. Especially, the modified VGG16, VGG19, MobileNet, and InceptionV3 models achieved 98%, 96%, 85%, and 75% accuracy respectively. This emphasizes the significance of leveraging pre-trained models through transfer learning techniques, which can substantially enhance the performance of the fire detection classification task, even when using a small dataset.