Monitoring via Video a Deep Convolutional Neural Network for Identifying Wildfires

Abstract

Wildfire monitoring has grown in importance, and vision-based fire detection technologies play a key role in this. Due to the rapid destruction of economic values and public safety that forest fires can wreak, wildfire warning systems are garnering increased interest. To lessen the impact of wildfires, a dark convolutional neural network (CNN)—a relatively new technology in image processing and video surveillance—is crucial. The original observation system is unable to apply Dark CNN Networks-based fire detection due to the high computational and memory requirements for identifying wildfires. We offer a computationally efficient and effective design for Dark Convolutional Neural Networks (CNNs) for wildfire detection, localization, and semantic understanding of the precise location of the fire. Here, we put forth a novel approach for picture recognition and classification based on Super pixels. It reduces computing requirements by making use of more convolutional segments and by omitting dense, fully connected layers. Our experimental setup proves that, mostly as a result of its greater depth, our suggested solution outperforms other, more complicated models in terms of accuracy.