A fire detection algorithm using convolutional neural network

Abstract EN

Fire destroys everything on its way.

It is the most dangerous hazard that causes disasters.

It can be started from a small ignition which could lead to a big loss or unwanted disaster.

People lose their lives from fires.

According to National Fire Protection Association (NFPA), the reported fire cases were nearly 1,400,000 in 2020 while those cases caused almost 3,500 civilian deaths.

In addition, the number of civilians injured from fires is around 15,000 and the estimated properties damage is around 21 billion in US Dollars.

Thus, detection of fire has become a very important topic especially due to the rapid and evolving of technology.

In this paper, a simple, fast and accurate method of fire detection using deep learning is proposed.

This method is developed based on image processing techniques.

If the fire is not detected early, then the Oxygen level decreases which leads to suffocation.

So, this method can help us by detecting fires at an early stage.

This approach filters an image into pixels based on thresholds according to some features such as colors, immobility source and flame texture with its reflection.

MATLAB is used as a simulation tool to conduct several experiments to verify the effectiveness of the proposed method.

The obtained results show that the accuracy was over 97% when applied on more than 700 images for training and testing purposes.