earticle

영어

The technologies underlying fire and smoke detection systems play crucial roles in ensuring that these systems deliver optimal performance in modern. In fact, fire can cause significant damage to lives and properties. In majority of cities, camera-monitoring systems have been already installed, to take advantage of availability of these kinds of systems encourage us to develop a cost-effective vision detection methods. However, this is a complex vision task by the reason of perspective deformations, unusual camera angles and viewpoints, and seasonal changes. To overcome these limitations, we propose a method-based on deep learning that uses a convolutional neural network, which employs dilated convolutions. We evaluated our method, by training and testing it on our custom-built dataset. Consisting of a collection of fire and smoke images that we collected and labeled manually. The performances of methods proposed in previous studies were compared with those of well-known state-of-the-art architectures; our experimental results indicate that the classification performance and complexity of our method was superior to those of previous methods. In addition, our method is designed to be well generalized for unseen data, that it offers effective generalization and also reduces the number of false alarms.