Video Surveillance Wildfire Detection using Dark Convolutional Neural Network

Abstract

Vision based fire detection systems have become important for wildfire surveillance. Wildfire detection systems are obtaining more attention because forest fires cause significant damage to both economic properties and public safety within a very short span of time. So it is essential to use Dark convolutional neural network to minimize the consequence of wildfire where Dark-CNN has become an emerging technology in image processing and video surveillance. The limitation with Dark CNN Networks-based fire detection is the usage in original observation system, because of the high memory and computational necessities for detecting wildfire. We propose an effective and computationally efficient Dark CNN Networks architecture, wildfire detection, localization, and semantic understanding of exact place where the fire occurs. Here we proposed a new algorithm using Super pixel based image classification and recognition. It utilizes more convolutional segments and contains no dense, complete connected layers, which avails keep the computational requisites to a minimum .Our experimental system illustrates that our proposed solution achieves more accuracy than others, more complex models, mainly due to its increased depth.